2009-07-08 23:01:31 +03:00
|
|
|
/*
|
|
|
|
* AAC coefficients encoder
|
|
|
|
* Copyright (C) 2008-2009 Konstantin Shishkov
|
|
|
|
*
|
|
|
|
* This file is part of FFmpeg.
|
|
|
|
*
|
|
|
|
* FFmpeg is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of the GNU Lesser General Public
|
|
|
|
* License as published by the Free Software Foundation; either
|
|
|
|
* version 2.1 of the License, or (at your option) any later version.
|
|
|
|
*
|
|
|
|
* FFmpeg is distributed in the hope that it will be useful,
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
* Lesser General Public License for more details.
|
|
|
|
*
|
|
|
|
* You should have received a copy of the GNU Lesser General Public
|
|
|
|
* License along with FFmpeg; if not, write to the Free Software
|
|
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
|
|
*/
|
|
|
|
|
|
|
|
/**
|
2010-04-20 17:45:34 +03:00
|
|
|
* @file
|
2009-07-08 23:01:31 +03:00
|
|
|
* AAC coefficients encoder
|
|
|
|
*/
|
|
|
|
|
|
|
|
/***********************************
|
|
|
|
* TODOs:
|
|
|
|
* speedup quantizer selection
|
|
|
|
* add sane pulse detection
|
|
|
|
***********************************/
|
|
|
|
|
2011-05-11 14:51:11 +03:00
|
|
|
#include "libavutil/libm.h" // brought forward to work around cygwin header breakage
|
|
|
|
|
2010-05-25 21:28:18 +03:00
|
|
|
#include <float.h>
|
2011-07-17 13:19:35 +03:00
|
|
|
#include "libavutil/mathematics.h"
|
2009-07-08 23:01:31 +03:00
|
|
|
#include "avcodec.h"
|
|
|
|
#include "put_bits.h"
|
|
|
|
#include "aac.h"
|
|
|
|
#include "aacenc.h"
|
|
|
|
#include "aactab.h"
|
|
|
|
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
/** Frequency in Hz for lower limit of noise substitution **/
|
|
|
|
#define NOISE_LOW_LIMIT 4000
|
|
|
|
|
|
|
|
/** Total number of usable codebooks **/
|
|
|
|
#define CB_TOT 13
|
|
|
|
|
2009-07-08 23:01:31 +03:00
|
|
|
/** bits needed to code codebook run value for long windows */
|
|
|
|
static const uint8_t run_value_bits_long[64] = {
|
|
|
|
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
|
|
|
|
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 10,
|
|
|
|
10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
|
|
|
|
10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 15
|
|
|
|
};
|
|
|
|
|
|
|
|
/** bits needed to code codebook run value for short windows */
|
|
|
|
static const uint8_t run_value_bits_short[16] = {
|
|
|
|
3, 3, 3, 3, 3, 3, 3, 6, 6, 6, 6, 6, 6, 6, 6, 9
|
|
|
|
};
|
|
|
|
|
2014-08-29 01:32:32 +03:00
|
|
|
static const uint8_t * const run_value_bits[2] = {
|
2009-07-08 23:01:31 +03:00
|
|
|
run_value_bits_long, run_value_bits_short
|
|
|
|
};
|
|
|
|
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
/** Map to convert values from BandCodingPath index to a codebook index **/
|
|
|
|
static const uint8_t aac_cb_out_map[CB_TOT] = {0,1,2,3,4,5,6,7,8,9,10,11,13};
|
|
|
|
/** Inverse map to convert from codebooks to BandCodingPath indices **/
|
|
|
|
static const uint8_t aac_cb_in_map[CB_TOT+1] = {0,1,2,3,4,5,6,7,8,9,10,11,0,12};
|
2009-07-08 23:01:31 +03:00
|
|
|
|
|
|
|
/**
|
|
|
|
* Quantize one coefficient.
|
|
|
|
* @return absolute value of the quantized coefficient
|
|
|
|
* @see 3GPP TS26.403 5.6.2 "Scalefactor determination"
|
|
|
|
*/
|
|
|
|
static av_always_inline int quant(float coef, const float Q)
|
|
|
|
{
|
2009-07-17 17:21:49 +03:00
|
|
|
float a = coef * Q;
|
|
|
|
return sqrtf(a * sqrtf(a)) + 0.4054;
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
|
2010-05-14 19:49:40 +03:00
|
|
|
static void quantize_bands(int *out, const float *in, const float *scaled,
|
2009-07-09 00:16:06 +03:00
|
|
|
int size, float Q34, int is_signed, int maxval)
|
2009-07-08 23:01:31 +03:00
|
|
|
{
|
|
|
|
int i;
|
|
|
|
double qc;
|
|
|
|
for (i = 0; i < size; i++) {
|
|
|
|
qc = scaled[i] * Q34;
|
2010-05-14 19:49:40 +03:00
|
|
|
out[i] = (int)FFMIN(qc + 0.4054, (double)maxval);
|
2009-07-08 23:01:31 +03:00
|
|
|
if (is_signed && in[i] < 0.0f) {
|
2010-05-14 19:49:40 +03:00
|
|
|
out[i] = -out[i];
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2009-07-09 00:16:06 +03:00
|
|
|
static void abs_pow34_v(float *out, const float *in, const int size)
|
2009-07-08 23:01:31 +03:00
|
|
|
{
|
|
|
|
#ifndef USE_REALLY_FULL_SEARCH
|
|
|
|
int i;
|
2009-07-17 17:21:49 +03:00
|
|
|
for (i = 0; i < size; i++) {
|
|
|
|
float a = fabsf(in[i]);
|
|
|
|
out[i] = sqrtf(a * sqrtf(a));
|
|
|
|
}
|
2009-07-08 23:01:31 +03:00
|
|
|
#endif /* USE_REALLY_FULL_SEARCH */
|
|
|
|
}
|
|
|
|
|
|
|
|
static const uint8_t aac_cb_range [12] = {0, 3, 3, 3, 3, 9, 9, 8, 8, 13, 13, 17};
|
|
|
|
static const uint8_t aac_cb_maxval[12] = {0, 1, 1, 2, 2, 4, 4, 7, 7, 12, 12, 16};
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Calculate rate distortion cost for quantizing with given codebook
|
|
|
|
*
|
|
|
|
* @return quantization distortion
|
|
|
|
*/
|
2010-07-16 23:02:46 +03:00
|
|
|
static av_always_inline float quantize_and_encode_band_cost_template(
|
|
|
|
struct AACEncContext *s,
|
2010-03-24 19:09:21 +02:00
|
|
|
PutBitContext *pb, const float *in,
|
2009-07-09 00:16:06 +03:00
|
|
|
const float *scaled, int size, int scale_idx,
|
|
|
|
int cb, const float lambda, const float uplim,
|
2010-07-16 23:02:46 +03:00
|
|
|
int *bits, int BT_ZERO, int BT_UNSIGNED,
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
int BT_PAIR, int BT_ESC, int BT_NOISE)
|
2009-07-08 23:01:31 +03:00
|
|
|
{
|
2012-01-07 00:24:37 +03:00
|
|
|
const int q_idx = POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512;
|
|
|
|
const float Q = ff_aac_pow2sf_tab [q_idx];
|
|
|
|
const float Q34 = ff_aac_pow34sf_tab[q_idx];
|
|
|
|
const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
|
2009-07-08 23:01:31 +03:00
|
|
|
const float CLIPPED_ESCAPE = 165140.0f*IQ;
|
2011-03-31 16:41:53 +03:00
|
|
|
int i, j;
|
2009-07-08 23:01:31 +03:00
|
|
|
float cost = 0;
|
2010-07-16 23:02:46 +03:00
|
|
|
const int dim = BT_PAIR ? 2 : 4;
|
2009-07-08 23:01:31 +03:00
|
|
|
int resbits = 0;
|
2010-05-14 19:49:40 +03:00
|
|
|
int off;
|
2009-07-08 23:01:31 +03:00
|
|
|
|
2010-07-16 23:02:46 +03:00
|
|
|
if (BT_ZERO) {
|
2009-07-08 23:36:45 +03:00
|
|
|
for (i = 0; i < size; i++)
|
2009-09-21 06:50:59 +03:00
|
|
|
cost += in[i]*in[i];
|
2009-07-17 01:20:17 +03:00
|
|
|
if (bits)
|
|
|
|
*bits = 0;
|
2009-09-21 06:50:59 +03:00
|
|
|
return cost * lambda;
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
if (BT_NOISE) {
|
|
|
|
for (i = 0; i < size; i++)
|
|
|
|
cost += in[i]*in[i];
|
|
|
|
if (bits)
|
|
|
|
*bits = 0;
|
|
|
|
return cost * lambda;
|
|
|
|
}
|
2010-03-24 19:09:21 +02:00
|
|
|
if (!scaled) {
|
|
|
|
abs_pow34_v(s->scoefs, in, size);
|
|
|
|
scaled = s->scoefs;
|
|
|
|
}
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
quantize_bands(s->qcoefs, in, scaled, size, Q34, !BT_UNSIGNED, aac_cb_maxval[cb]);
|
2010-07-16 23:02:46 +03:00
|
|
|
if (BT_UNSIGNED) {
|
2010-05-14 19:49:40 +03:00
|
|
|
off = 0;
|
|
|
|
} else {
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
off = aac_cb_maxval[cb];
|
2010-05-14 19:49:40 +03:00
|
|
|
}
|
2009-07-08 23:36:45 +03:00
|
|
|
for (i = 0; i < size; i += dim) {
|
2009-07-08 23:01:31 +03:00
|
|
|
const float *vec;
|
2010-05-14 19:49:40 +03:00
|
|
|
int *quants = s->qcoefs + i;
|
|
|
|
int curidx = 0;
|
|
|
|
int curbits;
|
|
|
|
float rd = 0.0f;
|
|
|
|
for (j = 0; j < dim; j++) {
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
curidx *= aac_cb_range[cb];
|
2010-05-14 19:49:40 +03:00
|
|
|
curidx += quants[j] + off;
|
|
|
|
}
|
2011-03-17 05:35:39 +02:00
|
|
|
curbits = ff_aac_spectral_bits[cb-1][curidx];
|
|
|
|
vec = &ff_aac_codebook_vectors[cb-1][curidx*dim];
|
|
|
|
if (BT_UNSIGNED) {
|
2011-03-31 16:41:53 +03:00
|
|
|
for (j = 0; j < dim; j++) {
|
|
|
|
float t = fabsf(in[i+j]);
|
2011-03-17 05:35:39 +02:00
|
|
|
float di;
|
2011-03-31 16:41:53 +03:00
|
|
|
if (BT_ESC && vec[j] == 64.0f) { //FIXME: slow
|
2011-03-17 05:35:39 +02:00
|
|
|
if (t >= CLIPPED_ESCAPE) {
|
|
|
|
di = t - CLIPPED_ESCAPE;
|
|
|
|
curbits += 21;
|
2009-07-08 23:36:45 +03:00
|
|
|
} else {
|
2015-02-20 02:35:35 +02:00
|
|
|
int c = av_clip_uintp2(quant(t, Q), 13);
|
2011-03-17 05:35:39 +02:00
|
|
|
di = t - c*cbrtf(c)*IQ;
|
|
|
|
curbits += av_log2(c)*2 - 4 + 1;
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
2011-03-17 05:35:39 +02:00
|
|
|
} else {
|
2011-03-31 16:41:53 +03:00
|
|
|
di = t - vec[j]*IQ;
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
2011-03-31 16:41:53 +03:00
|
|
|
if (vec[j] != 0.0f)
|
2011-03-17 05:35:39 +02:00
|
|
|
curbits++;
|
|
|
|
rd += di*di;
|
|
|
|
}
|
|
|
|
} else {
|
2011-03-31 16:41:53 +03:00
|
|
|
for (j = 0; j < dim; j++) {
|
|
|
|
float di = in[i+j] - vec[j]*IQ;
|
2011-03-17 05:35:39 +02:00
|
|
|
rd += di*di;
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
2011-03-17 05:35:39 +02:00
|
|
|
}
|
2010-05-14 19:49:40 +03:00
|
|
|
cost += rd * lambda + curbits;
|
|
|
|
resbits += curbits;
|
2009-07-08 23:36:45 +03:00
|
|
|
if (cost >= uplim)
|
2009-07-08 23:01:31 +03:00
|
|
|
return uplim;
|
2010-03-24 19:09:21 +02:00
|
|
|
if (pb) {
|
2010-07-19 21:37:45 +03:00
|
|
|
put_bits(pb, ff_aac_spectral_bits[cb-1][curidx], ff_aac_spectral_codes[cb-1][curidx]);
|
|
|
|
if (BT_UNSIGNED)
|
|
|
|
for (j = 0; j < dim; j++)
|
|
|
|
if (ff_aac_codebook_vectors[cb-1][curidx*dim+j] != 0.0f)
|
|
|
|
put_bits(pb, 1, in[i+j] < 0.0f);
|
|
|
|
if (BT_ESC) {
|
|
|
|
for (j = 0; j < 2; j++) {
|
|
|
|
if (ff_aac_codebook_vectors[cb-1][curidx*2+j] == 64.0f) {
|
2015-02-20 02:35:35 +02:00
|
|
|
int coef = av_clip_uintp2(quant(fabsf(in[i+j]), Q), 13);
|
2010-07-19 21:37:45 +03:00
|
|
|
int len = av_log2(coef);
|
|
|
|
|
|
|
|
put_bits(pb, len - 4 + 1, (1 << (len - 4 + 1)) - 2);
|
|
|
|
put_bits(pb, len, coef & ((1 << len) - 1));
|
|
|
|
}
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2010-03-24 19:09:21 +02:00
|
|
|
|
|
|
|
if (bits)
|
|
|
|
*bits = resbits;
|
|
|
|
return cost;
|
|
|
|
}
|
2010-07-16 23:02:46 +03:00
|
|
|
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
static float quantize_and_encode_band_cost_NONE(struct AACEncContext *s, PutBitContext *pb,
|
|
|
|
const float *in, const float *scaled,
|
|
|
|
int size, int scale_idx, int cb,
|
|
|
|
const float lambda, const float uplim,
|
|
|
|
int *bits) {
|
|
|
|
av_assert0(0);
|
|
|
|
return 0.0f;
|
|
|
|
}
|
|
|
|
|
|
|
|
#define QUANTIZE_AND_ENCODE_BAND_COST_FUNC(NAME, BT_ZERO, BT_UNSIGNED, BT_PAIR, BT_ESC, BT_NOISE) \
|
|
|
|
static float quantize_and_encode_band_cost_ ## NAME( \
|
2010-07-16 23:02:46 +03:00
|
|
|
struct AACEncContext *s, \
|
|
|
|
PutBitContext *pb, const float *in, \
|
|
|
|
const float *scaled, int size, int scale_idx, \
|
|
|
|
int cb, const float lambda, const float uplim, \
|
|
|
|
int *bits) { \
|
|
|
|
return quantize_and_encode_band_cost_template( \
|
|
|
|
s, pb, in, scaled, size, scale_idx, \
|
|
|
|
BT_ESC ? ESC_BT : cb, lambda, uplim, bits, \
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
BT_ZERO, BT_UNSIGNED, BT_PAIR, BT_ESC, BT_NOISE); \
|
2010-07-16 23:02:46 +03:00
|
|
|
}
|
|
|
|
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
QUANTIZE_AND_ENCODE_BAND_COST_FUNC(ZERO, 1, 0, 0, 0, 0)
|
|
|
|
QUANTIZE_AND_ENCODE_BAND_COST_FUNC(SQUAD, 0, 0, 0, 0, 0)
|
|
|
|
QUANTIZE_AND_ENCODE_BAND_COST_FUNC(UQUAD, 0, 1, 0, 0, 0)
|
|
|
|
QUANTIZE_AND_ENCODE_BAND_COST_FUNC(SPAIR, 0, 0, 1, 0, 0)
|
|
|
|
QUANTIZE_AND_ENCODE_BAND_COST_FUNC(UPAIR, 0, 1, 1, 0, 0)
|
|
|
|
QUANTIZE_AND_ENCODE_BAND_COST_FUNC(ESC, 0, 1, 1, 1, 0)
|
|
|
|
QUANTIZE_AND_ENCODE_BAND_COST_FUNC(NOISE, 0, 0, 0, 0, 1)
|
2010-07-16 23:02:46 +03:00
|
|
|
|
2010-07-16 23:28:55 +03:00
|
|
|
static float (*const quantize_and_encode_band_cost_arr[])(
|
2010-07-16 23:02:46 +03:00
|
|
|
struct AACEncContext *s,
|
|
|
|
PutBitContext *pb, const float *in,
|
|
|
|
const float *scaled, int size, int scale_idx,
|
|
|
|
int cb, const float lambda, const float uplim,
|
|
|
|
int *bits) = {
|
|
|
|
quantize_and_encode_band_cost_ZERO,
|
|
|
|
quantize_and_encode_band_cost_SQUAD,
|
|
|
|
quantize_and_encode_band_cost_SQUAD,
|
|
|
|
quantize_and_encode_band_cost_UQUAD,
|
|
|
|
quantize_and_encode_band_cost_UQUAD,
|
|
|
|
quantize_and_encode_band_cost_SPAIR,
|
|
|
|
quantize_and_encode_band_cost_SPAIR,
|
|
|
|
quantize_and_encode_band_cost_UPAIR,
|
|
|
|
quantize_and_encode_band_cost_UPAIR,
|
|
|
|
quantize_and_encode_band_cost_UPAIR,
|
|
|
|
quantize_and_encode_band_cost_UPAIR,
|
|
|
|
quantize_and_encode_band_cost_ESC,
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
quantize_and_encode_band_cost_NONE, /* CB 12 doesn't exist */
|
|
|
|
quantize_and_encode_band_cost_NOISE,
|
2010-07-16 23:02:46 +03:00
|
|
|
};
|
|
|
|
|
|
|
|
#define quantize_and_encode_band_cost( \
|
|
|
|
s, pb, in, scaled, size, scale_idx, cb, \
|
|
|
|
lambda, uplim, bits) \
|
|
|
|
quantize_and_encode_band_cost_arr[cb]( \
|
|
|
|
s, pb, in, scaled, size, scale_idx, cb, \
|
|
|
|
lambda, uplim, bits)
|
|
|
|
|
2010-03-24 19:09:21 +02:00
|
|
|
static float quantize_band_cost(struct AACEncContext *s, const float *in,
|
|
|
|
const float *scaled, int size, int scale_idx,
|
|
|
|
int cb, const float lambda, const float uplim,
|
|
|
|
int *bits)
|
|
|
|
{
|
|
|
|
return quantize_and_encode_band_cost(s, NULL, in, scaled, size, scale_idx,
|
|
|
|
cb, lambda, uplim, bits);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void quantize_and_encode_band(struct AACEncContext *s, PutBitContext *pb,
|
|
|
|
const float *in, int size, int scale_idx,
|
|
|
|
int cb, const float lambda)
|
|
|
|
{
|
|
|
|
quantize_and_encode_band_cost(s, pb, in, NULL, size, scale_idx, cb, lambda,
|
|
|
|
INFINITY, NULL);
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
|
2010-05-25 21:35:01 +03:00
|
|
|
static float find_max_val(int group_len, int swb_size, const float *scaled) {
|
2010-05-25 21:24:16 +03:00
|
|
|
float maxval = 0.0f;
|
2010-05-25 21:35:01 +03:00
|
|
|
int w2, i;
|
2010-05-25 21:24:16 +03:00
|
|
|
for (w2 = 0; w2 < group_len; w2++) {
|
|
|
|
for (i = 0; i < swb_size; i++) {
|
|
|
|
maxval = FFMAX(maxval, scaled[w2*128+i]);
|
|
|
|
}
|
|
|
|
}
|
2010-05-25 21:35:01 +03:00
|
|
|
return maxval;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int find_min_book(float maxval, int sf) {
|
2011-04-26 21:45:48 +03:00
|
|
|
float Q = ff_aac_pow2sf_tab[POW_SF2_ZERO - sf + SCALE_ONE_POS - SCALE_DIV_512];
|
2010-05-25 21:35:01 +03:00
|
|
|
float Q34 = sqrtf(Q * sqrtf(Q));
|
|
|
|
int qmaxval, cb;
|
2010-05-25 21:24:16 +03:00
|
|
|
qmaxval = maxval * Q34 + 0.4054f;
|
|
|
|
if (qmaxval == 0) cb = 0;
|
|
|
|
else if (qmaxval == 1) cb = 1;
|
|
|
|
else if (qmaxval == 2) cb = 3;
|
|
|
|
else if (qmaxval <= 4) cb = 5;
|
|
|
|
else if (qmaxval <= 7) cb = 7;
|
|
|
|
else if (qmaxval <= 12) cb = 9;
|
|
|
|
else cb = 11;
|
|
|
|
return cb;
|
|
|
|
}
|
|
|
|
|
2009-07-08 23:01:31 +03:00
|
|
|
/**
|
|
|
|
* structure used in optimal codebook search
|
|
|
|
*/
|
|
|
|
typedef struct BandCodingPath {
|
|
|
|
int prev_idx; ///< pointer to the previous path point
|
|
|
|
float cost; ///< path cost
|
|
|
|
int run;
|
|
|
|
} BandCodingPath;
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Encode band info for single window group bands.
|
|
|
|
*/
|
|
|
|
static void encode_window_bands_info(AACEncContext *s, SingleChannelElement *sce,
|
|
|
|
int win, int group_len, const float lambda)
|
|
|
|
{
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
BandCodingPath path[120][CB_TOT];
|
2011-06-01 19:26:27 +03:00
|
|
|
int w, swb, cb, start, size;
|
2009-07-08 23:01:31 +03:00
|
|
|
int i, j;
|
2009-07-09 00:16:06 +03:00
|
|
|
const int max_sfb = sce->ics.max_sfb;
|
2009-07-08 23:01:31 +03:00
|
|
|
const int run_bits = sce->ics.num_windows == 1 ? 5 : 3;
|
2009-07-09 00:16:06 +03:00
|
|
|
const int run_esc = (1 << run_bits) - 1;
|
2009-07-08 23:01:31 +03:00
|
|
|
int idx, ppos, count;
|
|
|
|
int stackrun[120], stackcb[120], stack_len;
|
|
|
|
float next_minrd = INFINITY;
|
|
|
|
int next_mincb = 0;
|
|
|
|
|
|
|
|
abs_pow34_v(s->scoefs, sce->coeffs, 1024);
|
|
|
|
start = win*128;
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
for (cb = 0; cb < CB_TOT; cb++) {
|
2009-07-09 00:16:06 +03:00
|
|
|
path[0][cb].cost = 0.0f;
|
2009-07-08 23:01:31 +03:00
|
|
|
path[0][cb].prev_idx = -1;
|
2009-07-09 00:16:06 +03:00
|
|
|
path[0][cb].run = 0;
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
2009-07-08 23:36:45 +03:00
|
|
|
for (swb = 0; swb < max_sfb; swb++) {
|
2009-07-08 23:01:31 +03:00
|
|
|
size = sce->ics.swb_sizes[swb];
|
2009-07-08 23:36:45 +03:00
|
|
|
if (sce->zeroes[win*16 + swb]) {
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
for (cb = 0; cb < CB_TOT; cb++) {
|
2009-07-08 23:01:31 +03:00
|
|
|
path[swb+1][cb].prev_idx = cb;
|
2009-07-09 00:16:06 +03:00
|
|
|
path[swb+1][cb].cost = path[swb][cb].cost;
|
|
|
|
path[swb+1][cb].run = path[swb][cb].run + 1;
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
2009-07-08 23:36:45 +03:00
|
|
|
} else {
|
2009-07-08 23:01:31 +03:00
|
|
|
float minrd = next_minrd;
|
|
|
|
int mincb = next_mincb;
|
|
|
|
next_minrd = INFINITY;
|
|
|
|
next_mincb = 0;
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
for (cb = 0; cb < CB_TOT; cb++) {
|
2009-07-08 23:01:31 +03:00
|
|
|
float cost_stay_here, cost_get_here;
|
|
|
|
float rd = 0.0f;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w = 0; w < group_len; w++) {
|
2011-06-15 11:50:25 +03:00
|
|
|
FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[(win+w)*16+swb];
|
2009-07-08 23:01:31 +03:00
|
|
|
rd += quantize_band_cost(s, sce->coeffs + start + w*128,
|
|
|
|
s->scoefs + start + w*128, size,
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
sce->sf_idx[(win+w)*16+swb], aac_cb_out_map[cb],
|
2009-07-08 23:01:31 +03:00
|
|
|
lambda / band->threshold, INFINITY, NULL);
|
|
|
|
}
|
|
|
|
cost_stay_here = path[swb][cb].cost + rd;
|
|
|
|
cost_get_here = minrd + rd + run_bits + 4;
|
2009-07-08 23:36:45 +03:00
|
|
|
if ( run_value_bits[sce->ics.num_windows == 8][path[swb][cb].run]
|
2009-07-09 00:16:06 +03:00
|
|
|
!= run_value_bits[sce->ics.num_windows == 8][path[swb][cb].run+1])
|
2009-07-08 23:01:31 +03:00
|
|
|
cost_stay_here += run_bits;
|
|
|
|
if (cost_get_here < cost_stay_here) {
|
|
|
|
path[swb+1][cb].prev_idx = mincb;
|
|
|
|
path[swb+1][cb].cost = cost_get_here;
|
|
|
|
path[swb+1][cb].run = 1;
|
|
|
|
} else {
|
|
|
|
path[swb+1][cb].prev_idx = cb;
|
|
|
|
path[swb+1][cb].cost = cost_stay_here;
|
|
|
|
path[swb+1][cb].run = path[swb][cb].run + 1;
|
|
|
|
}
|
|
|
|
if (path[swb+1][cb].cost < next_minrd) {
|
|
|
|
next_minrd = path[swb+1][cb].cost;
|
|
|
|
next_mincb = cb;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
start += sce->ics.swb_sizes[swb];
|
|
|
|
}
|
|
|
|
|
|
|
|
//convert resulting path from backward-linked list
|
|
|
|
stack_len = 0;
|
2009-07-09 00:16:06 +03:00
|
|
|
idx = 0;
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
for (cb = 1; cb < CB_TOT; cb++)
|
2009-07-08 23:36:45 +03:00
|
|
|
if (path[max_sfb][cb].cost < path[max_sfb][idx].cost)
|
2009-07-08 23:01:31 +03:00
|
|
|
idx = cb;
|
|
|
|
ppos = max_sfb;
|
2009-07-09 00:16:06 +03:00
|
|
|
while (ppos > 0) {
|
2009-07-08 23:01:31 +03:00
|
|
|
cb = idx;
|
|
|
|
stackrun[stack_len] = path[ppos][cb].run;
|
|
|
|
stackcb [stack_len] = cb;
|
|
|
|
idx = path[ppos-path[ppos][cb].run+1][cb].prev_idx;
|
|
|
|
ppos -= path[ppos][cb].run;
|
|
|
|
stack_len++;
|
|
|
|
}
|
|
|
|
//perform actual band info encoding
|
|
|
|
start = 0;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (i = stack_len - 1; i >= 0; i--) {
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
cb = aac_cb_out_map[stackcb[i]];
|
|
|
|
put_bits(&s->pb, 4, cb);
|
2009-07-08 23:01:31 +03:00
|
|
|
count = stackrun[i];
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
memset(sce->zeroes + win*16 + start, !cb, count);
|
2009-07-08 23:01:31 +03:00
|
|
|
//XXX: memset when band_type is also uint8_t
|
2009-07-08 23:36:45 +03:00
|
|
|
for (j = 0; j < count; j++) {
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
sce->band_type[win*16 + start] = cb;
|
2009-07-08 23:01:31 +03:00
|
|
|
start++;
|
|
|
|
}
|
2009-07-09 00:16:06 +03:00
|
|
|
while (count >= run_esc) {
|
2009-07-08 23:01:31 +03:00
|
|
|
put_bits(&s->pb, run_bits, run_esc);
|
|
|
|
count -= run_esc;
|
|
|
|
}
|
|
|
|
put_bits(&s->pb, run_bits, count);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-05-14 19:49:51 +03:00
|
|
|
static void codebook_trellis_rate(AACEncContext *s, SingleChannelElement *sce,
|
|
|
|
int win, int group_len, const float lambda)
|
|
|
|
{
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
BandCodingPath path[120][CB_TOT];
|
2011-06-01 19:26:27 +03:00
|
|
|
int w, swb, cb, start, size;
|
2010-05-14 19:49:51 +03:00
|
|
|
int i, j;
|
|
|
|
const int max_sfb = sce->ics.max_sfb;
|
|
|
|
const int run_bits = sce->ics.num_windows == 1 ? 5 : 3;
|
|
|
|
const int run_esc = (1 << run_bits) - 1;
|
|
|
|
int idx, ppos, count;
|
|
|
|
int stackrun[120], stackcb[120], stack_len;
|
2012-01-09 04:38:39 +03:00
|
|
|
float next_minbits = INFINITY;
|
2010-05-14 19:49:51 +03:00
|
|
|
int next_mincb = 0;
|
|
|
|
|
|
|
|
abs_pow34_v(s->scoefs, sce->coeffs, 1024);
|
|
|
|
start = win*128;
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
for (cb = 0; cb < CB_TOT; cb++) {
|
2010-05-14 19:49:51 +03:00
|
|
|
path[0][cb].cost = run_bits+4;
|
|
|
|
path[0][cb].prev_idx = -1;
|
|
|
|
path[0][cb].run = 0;
|
|
|
|
}
|
|
|
|
for (swb = 0; swb < max_sfb; swb++) {
|
|
|
|
size = sce->ics.swb_sizes[swb];
|
|
|
|
if (sce->zeroes[win*16 + swb]) {
|
2011-06-22 09:12:42 +03:00
|
|
|
float cost_stay_here = path[swb][0].cost;
|
2012-01-09 04:38:39 +03:00
|
|
|
float cost_get_here = next_minbits + run_bits + 4;
|
2011-06-22 09:12:42 +03:00
|
|
|
if ( run_value_bits[sce->ics.num_windows == 8][path[swb][0].run]
|
|
|
|
!= run_value_bits[sce->ics.num_windows == 8][path[swb][0].run+1])
|
|
|
|
cost_stay_here += run_bits;
|
|
|
|
if (cost_get_here < cost_stay_here) {
|
|
|
|
path[swb+1][0].prev_idx = next_mincb;
|
|
|
|
path[swb+1][0].cost = cost_get_here;
|
|
|
|
path[swb+1][0].run = 1;
|
|
|
|
} else {
|
|
|
|
path[swb+1][0].prev_idx = 0;
|
|
|
|
path[swb+1][0].cost = cost_stay_here;
|
|
|
|
path[swb+1][0].run = path[swb][0].run + 1;
|
|
|
|
}
|
2012-01-09 04:38:39 +03:00
|
|
|
next_minbits = path[swb+1][0].cost;
|
2011-06-22 09:12:42 +03:00
|
|
|
next_mincb = 0;
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
for (cb = 1; cb < CB_TOT; cb++) {
|
2011-06-22 09:12:42 +03:00
|
|
|
path[swb+1][cb].cost = 61450;
|
|
|
|
path[swb+1][cb].prev_idx = -1;
|
|
|
|
path[swb+1][cb].run = 0;
|
2010-05-14 19:49:51 +03:00
|
|
|
}
|
|
|
|
} else {
|
2012-01-09 04:38:39 +03:00
|
|
|
float minbits = next_minbits;
|
2010-05-14 19:49:51 +03:00
|
|
|
int mincb = next_mincb;
|
|
|
|
int startcb = sce->band_type[win*16+swb];
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
startcb = aac_cb_in_map[startcb];
|
2012-01-09 04:38:39 +03:00
|
|
|
next_minbits = INFINITY;
|
2010-05-14 19:49:51 +03:00
|
|
|
next_mincb = 0;
|
|
|
|
for (cb = 0; cb < startcb; cb++) {
|
|
|
|
path[swb+1][cb].cost = 61450;
|
|
|
|
path[swb+1][cb].prev_idx = -1;
|
|
|
|
path[swb+1][cb].run = 0;
|
|
|
|
}
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
for (cb = startcb; cb < CB_TOT; cb++) {
|
2010-05-14 19:49:51 +03:00
|
|
|
float cost_stay_here, cost_get_here;
|
2012-01-09 04:38:39 +03:00
|
|
|
float bits = 0.0f;
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
if (cb == 12 && sce->band_type[win*16+swb] != NOISE_BT) {
|
|
|
|
path[swb+1][cb].cost = 61450;
|
|
|
|
path[swb+1][cb].prev_idx = -1;
|
|
|
|
path[swb+1][cb].run = 0;
|
|
|
|
continue;
|
|
|
|
}
|
2010-05-14 19:49:51 +03:00
|
|
|
for (w = 0; w < group_len; w++) {
|
2012-01-09 04:38:39 +03:00
|
|
|
bits += quantize_band_cost(s, sce->coeffs + start + w*128,
|
|
|
|
s->scoefs + start + w*128, size,
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
sce->sf_idx[(win+w)*16+swb],
|
|
|
|
aac_cb_out_map[cb],
|
2012-01-09 04:38:39 +03:00
|
|
|
0, INFINITY, NULL);
|
2010-05-14 19:49:51 +03:00
|
|
|
}
|
2012-01-09 04:38:39 +03:00
|
|
|
cost_stay_here = path[swb][cb].cost + bits;
|
|
|
|
cost_get_here = minbits + bits + run_bits + 4;
|
2010-05-14 19:49:51 +03:00
|
|
|
if ( run_value_bits[sce->ics.num_windows == 8][path[swb][cb].run]
|
|
|
|
!= run_value_bits[sce->ics.num_windows == 8][path[swb][cb].run+1])
|
|
|
|
cost_stay_here += run_bits;
|
|
|
|
if (cost_get_here < cost_stay_here) {
|
|
|
|
path[swb+1][cb].prev_idx = mincb;
|
|
|
|
path[swb+1][cb].cost = cost_get_here;
|
|
|
|
path[swb+1][cb].run = 1;
|
|
|
|
} else {
|
|
|
|
path[swb+1][cb].prev_idx = cb;
|
|
|
|
path[swb+1][cb].cost = cost_stay_here;
|
|
|
|
path[swb+1][cb].run = path[swb][cb].run + 1;
|
|
|
|
}
|
2012-01-09 04:38:39 +03:00
|
|
|
if (path[swb+1][cb].cost < next_minbits) {
|
|
|
|
next_minbits = path[swb+1][cb].cost;
|
2010-05-14 19:49:51 +03:00
|
|
|
next_mincb = cb;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
start += sce->ics.swb_sizes[swb];
|
|
|
|
}
|
|
|
|
|
|
|
|
//convert resulting path from backward-linked list
|
|
|
|
stack_len = 0;
|
|
|
|
idx = 0;
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
for (cb = 1; cb < CB_TOT; cb++)
|
2010-05-14 19:49:51 +03:00
|
|
|
if (path[max_sfb][cb].cost < path[max_sfb][idx].cost)
|
|
|
|
idx = cb;
|
|
|
|
ppos = max_sfb;
|
|
|
|
while (ppos > 0) {
|
2012-09-04 05:38:28 +03:00
|
|
|
av_assert1(idx >= 0);
|
2010-05-14 19:49:51 +03:00
|
|
|
cb = idx;
|
|
|
|
stackrun[stack_len] = path[ppos][cb].run;
|
|
|
|
stackcb [stack_len] = cb;
|
|
|
|
idx = path[ppos-path[ppos][cb].run+1][cb].prev_idx;
|
|
|
|
ppos -= path[ppos][cb].run;
|
|
|
|
stack_len++;
|
|
|
|
}
|
|
|
|
//perform actual band info encoding
|
|
|
|
start = 0;
|
|
|
|
for (i = stack_len - 1; i >= 0; i--) {
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
cb = aac_cb_out_map[stackcb[i]];
|
|
|
|
put_bits(&s->pb, 4, cb);
|
2010-05-14 19:49:51 +03:00
|
|
|
count = stackrun[i];
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
memset(sce->zeroes + win*16 + start, !cb, count);
|
2010-05-14 19:49:51 +03:00
|
|
|
//XXX: memset when band_type is also uint8_t
|
|
|
|
for (j = 0; j < count; j++) {
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
sce->band_type[win*16 + start] = cb;
|
2010-05-14 19:49:51 +03:00
|
|
|
start++;
|
|
|
|
}
|
|
|
|
while (count >= run_esc) {
|
|
|
|
put_bits(&s->pb, run_bits, run_esc);
|
|
|
|
count -= run_esc;
|
|
|
|
}
|
|
|
|
put_bits(&s->pb, run_bits, count);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-07-24 01:37:42 +03:00
|
|
|
/** Return the minimum scalefactor where the quantized coef does not clip. */
|
|
|
|
static av_always_inline uint8_t coef2minsf(float coef) {
|
2010-07-24 05:10:59 +03:00
|
|
|
return av_clip_uint8(log2f(coef)*4 - 69 + SCALE_ONE_POS - SCALE_DIV_512);
|
2010-07-24 01:37:42 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
/** Return the maximum scalefactor where the quantized coef is not zero. */
|
|
|
|
static av_always_inline uint8_t coef2maxsf(float coef) {
|
2010-07-24 05:10:59 +03:00
|
|
|
return av_clip_uint8(log2f(coef)*4 + 6 + SCALE_ONE_POS - SCALE_DIV_512);
|
2010-07-24 01:37:42 +03:00
|
|
|
}
|
|
|
|
|
2009-07-08 23:01:31 +03:00
|
|
|
typedef struct TrellisPath {
|
|
|
|
float cost;
|
|
|
|
int prev;
|
|
|
|
} TrellisPath;
|
|
|
|
|
2009-09-21 07:32:51 +03:00
|
|
|
#define TRELLIS_STAGES 121
|
2010-05-25 21:28:18 +03:00
|
|
|
#define TRELLIS_STATES (SCALE_MAX_DIFF+1)
|
2009-09-21 07:32:51 +03:00
|
|
|
|
2009-07-08 23:01:31 +03:00
|
|
|
static void search_for_quantizers_anmr(AVCodecContext *avctx, AACEncContext *s,
|
2009-07-09 00:16:06 +03:00
|
|
|
SingleChannelElement *sce,
|
|
|
|
const float lambda)
|
2009-07-08 23:01:31 +03:00
|
|
|
{
|
|
|
|
int q, w, w2, g, start = 0;
|
2009-09-21 07:15:12 +03:00
|
|
|
int i, j;
|
2009-07-08 23:01:31 +03:00
|
|
|
int idx;
|
2009-09-21 07:32:51 +03:00
|
|
|
TrellisPath paths[TRELLIS_STAGES][TRELLIS_STATES];
|
|
|
|
int bandaddr[TRELLIS_STAGES];
|
2009-07-08 23:01:31 +03:00
|
|
|
int minq;
|
|
|
|
float mincost;
|
2010-05-25 21:28:18 +03:00
|
|
|
float q0f = FLT_MAX, q1f = 0.0f, qnrgf = 0.0f;
|
|
|
|
int q0, q1, qcnt = 0;
|
|
|
|
|
|
|
|
for (i = 0; i < 1024; i++) {
|
|
|
|
float t = fabsf(sce->coeffs[i]);
|
|
|
|
if (t > 0.0f) {
|
|
|
|
q0f = FFMIN(q0f, t);
|
|
|
|
q1f = FFMAX(q1f, t);
|
|
|
|
qnrgf += t*t;
|
|
|
|
qcnt++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!qcnt) {
|
|
|
|
memset(sce->sf_idx, 0, sizeof(sce->sf_idx));
|
|
|
|
memset(sce->zeroes, 1, sizeof(sce->zeroes));
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
//minimum scalefactor index is when minimum nonzero coefficient after quantizing is not clipped
|
2010-07-24 01:37:42 +03:00
|
|
|
q0 = coef2minsf(q0f);
|
2010-05-25 21:28:18 +03:00
|
|
|
//maximum scalefactor index is when maximum coefficient after quantizing is still not zero
|
2010-07-24 01:37:42 +03:00
|
|
|
q1 = coef2maxsf(q1f);
|
2010-05-25 21:28:18 +03:00
|
|
|
if (q1 - q0 > 60) {
|
|
|
|
int q0low = q0;
|
|
|
|
int q1high = q1;
|
|
|
|
//minimum scalefactor index is when maximum nonzero coefficient after quantizing is not clipped
|
2010-07-24 05:10:59 +03:00
|
|
|
int qnrg = av_clip_uint8(log2f(sqrtf(qnrgf/qcnt))*4 - 31 + SCALE_ONE_POS - SCALE_DIV_512);
|
2010-05-25 21:28:18 +03:00
|
|
|
q1 = qnrg + 30;
|
|
|
|
q0 = qnrg - 30;
|
|
|
|
if (q0 < q0low) {
|
|
|
|
q1 += q0low - q0;
|
|
|
|
q0 = q0low;
|
|
|
|
} else if (q1 > q1high) {
|
|
|
|
q0 -= q1 - q1high;
|
|
|
|
q1 = q1high;
|
|
|
|
}
|
|
|
|
}
|
2009-07-08 23:01:31 +03:00
|
|
|
|
2009-09-21 07:32:51 +03:00
|
|
|
for (i = 0; i < TRELLIS_STATES; i++) {
|
2009-09-21 07:15:12 +03:00
|
|
|
paths[0][i].cost = 0.0f;
|
|
|
|
paths[0][i].prev = -1;
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
2009-09-21 07:32:51 +03:00
|
|
|
for (j = 1; j < TRELLIS_STAGES; j++) {
|
|
|
|
for (i = 0; i < TRELLIS_STATES; i++) {
|
2009-09-21 07:15:12 +03:00
|
|
|
paths[j][i].cost = INFINITY;
|
|
|
|
paths[j][i].prev = -2;
|
|
|
|
}
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
2009-09-21 07:15:12 +03:00
|
|
|
idx = 1;
|
2009-07-08 23:01:31 +03:00
|
|
|
abs_pow34_v(s->scoefs, sce->coeffs, 1024);
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
|
2009-07-08 23:01:31 +03:00
|
|
|
start = w*128;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (g = 0; g < sce->ics.num_swb; g++) {
|
2009-07-08 23:01:31 +03:00
|
|
|
const float *coefs = sce->coeffs + start;
|
|
|
|
float qmin, qmax;
|
|
|
|
int nz = 0;
|
|
|
|
|
2009-09-21 07:15:12 +03:00
|
|
|
bandaddr[idx] = w * 16 + g;
|
2009-07-08 23:01:31 +03:00
|
|
|
qmin = INT_MAX;
|
|
|
|
qmax = 0.0f;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
|
2011-06-15 11:50:25 +03:00
|
|
|
FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
|
2009-07-08 23:36:45 +03:00
|
|
|
if (band->energy <= band->threshold || band->threshold == 0.0f) {
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->zeroes[(w+w2)*16+g] = 1;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
sce->zeroes[(w+w2)*16+g] = 0;
|
|
|
|
nz = 1;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (i = 0; i < sce->ics.swb_sizes[g]; i++) {
|
2009-07-08 23:01:31 +03:00
|
|
|
float t = fabsf(coefs[w2*128+i]);
|
2009-07-09 00:36:33 +03:00
|
|
|
if (t > 0.0f)
|
2009-07-09 20:22:59 +03:00
|
|
|
qmin = FFMIN(qmin, t);
|
|
|
|
qmax = FFMAX(qmax, t);
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
}
|
2009-07-08 23:36:45 +03:00
|
|
|
if (nz) {
|
2009-07-08 23:01:31 +03:00
|
|
|
int minscale, maxscale;
|
|
|
|
float minrd = INFINITY;
|
2010-05-26 02:25:29 +03:00
|
|
|
float maxval;
|
2009-07-08 23:01:31 +03:00
|
|
|
//minimum scalefactor index is when minimum nonzero coefficient after quantizing is not clipped
|
2010-07-24 01:37:42 +03:00
|
|
|
minscale = coef2minsf(qmin);
|
2009-07-08 23:01:31 +03:00
|
|
|
//maximum scalefactor index is when maximum coefficient after quantizing is still not zero
|
2010-07-24 01:37:42 +03:00
|
|
|
maxscale = coef2maxsf(qmax);
|
2010-05-25 21:28:18 +03:00
|
|
|
minscale = av_clip(minscale - q0, 0, TRELLIS_STATES - 1);
|
|
|
|
maxscale = av_clip(maxscale - q0, 0, TRELLIS_STATES);
|
2010-05-26 02:25:29 +03:00
|
|
|
maxval = find_max_val(sce->ics.group_len[w], sce->ics.swb_sizes[g], s->scoefs+start);
|
2009-07-08 23:36:45 +03:00
|
|
|
for (q = minscale; q < maxscale; q++) {
|
2010-05-25 21:31:55 +03:00
|
|
|
float dist = 0;
|
2010-05-25 21:35:01 +03:00
|
|
|
int cb = find_min_book(maxval, sce->sf_idx[w*16+g]);
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
|
2011-06-15 11:50:25 +03:00
|
|
|
FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
|
2010-05-25 21:31:55 +03:00
|
|
|
dist += quantize_band_cost(s, coefs + w2*128, s->scoefs + start + w2*128, sce->ics.swb_sizes[g],
|
2010-05-28 23:45:35 +03:00
|
|
|
q + q0, cb, lambda / band->threshold, INFINITY, NULL);
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
2009-07-09 20:22:59 +03:00
|
|
|
minrd = FFMIN(minrd, dist);
|
2009-07-08 23:01:31 +03:00
|
|
|
|
2010-05-25 21:28:18 +03:00
|
|
|
for (i = 0; i < q1 - q0; i++) {
|
2009-07-08 23:01:31 +03:00
|
|
|
float cost;
|
2009-09-21 07:15:12 +03:00
|
|
|
cost = paths[idx - 1][i].cost + dist
|
2009-07-08 23:01:31 +03:00
|
|
|
+ ff_aac_scalefactor_bits[q - i + SCALE_DIFF_ZERO];
|
2010-05-25 21:28:18 +03:00
|
|
|
if (cost < paths[idx][q].cost) {
|
2009-09-21 07:15:12 +03:00
|
|
|
paths[idx][q].cost = cost;
|
|
|
|
paths[idx][q].prev = i;
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2009-07-08 23:36:45 +03:00
|
|
|
} else {
|
2010-05-25 21:28:18 +03:00
|
|
|
for (q = 0; q < q1 - q0; q++) {
|
2010-05-28 23:45:35 +03:00
|
|
|
paths[idx][q].cost = paths[idx - 1][q].cost + 1;
|
|
|
|
paths[idx][q].prev = q;
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
sce->zeroes[w*16+g] = !nz;
|
|
|
|
start += sce->ics.swb_sizes[g];
|
2009-09-21 07:15:12 +03:00
|
|
|
idx++;
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
}
|
2009-09-21 07:15:12 +03:00
|
|
|
idx--;
|
|
|
|
mincost = paths[idx][0].cost;
|
|
|
|
minq = 0;
|
2009-09-21 07:32:51 +03:00
|
|
|
for (i = 1; i < TRELLIS_STATES; i++) {
|
2009-09-21 07:15:12 +03:00
|
|
|
if (paths[idx][i].cost < mincost) {
|
|
|
|
mincost = paths[idx][i].cost;
|
|
|
|
minq = i;
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
}
|
2009-09-21 07:15:12 +03:00
|
|
|
while (idx) {
|
2010-05-25 21:28:18 +03:00
|
|
|
sce->sf_idx[bandaddr[idx]] = minq + q0;
|
2009-09-21 07:15:12 +03:00
|
|
|
minq = paths[idx][minq].prev;
|
|
|
|
idx--;
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
//set the same quantizers inside window groups
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
|
|
|
|
for (g = 0; g < sce->ics.num_swb; g++)
|
|
|
|
for (w2 = 1; w2 < sce->ics.group_len[w]; w2++)
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->sf_idx[(w+w2)*16+g] = sce->sf_idx[w*16+g];
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* two-loop quantizers search taken from ISO 13818-7 Appendix C
|
|
|
|
*/
|
2009-07-09 00:16:06 +03:00
|
|
|
static void search_for_quantizers_twoloop(AVCodecContext *avctx,
|
|
|
|
AACEncContext *s,
|
|
|
|
SingleChannelElement *sce,
|
|
|
|
const float lambda)
|
2009-07-08 23:01:31 +03:00
|
|
|
{
|
|
|
|
int start = 0, i, w, w2, g;
|
2013-05-05 00:36:37 +03:00
|
|
|
int destbits = avctx->bit_rate * 1024.0 / avctx->sample_rate / avctx->channels * (lambda / 120.f);
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
const float freq_mult = avctx->sample_rate/(1024.0f/sce->ics.num_windows)/2.0f;
|
|
|
|
float dists[128] = { 0 }, uplims[128] = { 0 };
|
2010-07-24 20:11:51 +03:00
|
|
|
float maxvals[128];
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
int noise_sf[128] = { 0 };
|
|
|
|
int fflag, minscaler, minscaler_n;
|
2009-07-09 00:16:06 +03:00
|
|
|
int its = 0;
|
2009-07-08 23:01:31 +03:00
|
|
|
int allz = 0;
|
|
|
|
float minthr = INFINITY;
|
|
|
|
|
2012-04-06 16:25:05 +03:00
|
|
|
// for values above this the decoder might end up in an endless loop
|
|
|
|
// due to always having more bits than what can be encoded.
|
|
|
|
destbits = FFMIN(destbits, 5800);
|
2009-07-08 23:01:31 +03:00
|
|
|
//XXX: some heuristic to determine initial quantizers will reduce search time
|
|
|
|
//determine zero bands and upper limits
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
start = 0;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (g = 0; g < sce->ics.num_swb; g++) {
|
2009-07-08 23:01:31 +03:00
|
|
|
int nz = 0;
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
float uplim = 0.0f, energy = 0.0f;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
|
2011-06-15 11:50:25 +03:00
|
|
|
FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
|
2009-07-08 23:01:31 +03:00
|
|
|
uplim += band->threshold;
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
energy += band->energy;
|
2009-07-08 23:36:45 +03:00
|
|
|
if (band->energy <= band->threshold || band->threshold == 0.0f) {
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->zeroes[(w+w2)*16+g] = 1;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
nz = 1;
|
|
|
|
}
|
|
|
|
uplims[w*16+g] = uplim *512;
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
if (s->options.pns && start*freq_mult > NOISE_LOW_LIMIT && energy < uplim * 1.2f) {
|
|
|
|
noise_sf[w*16+g] = av_clip(4+FFMIN(log2f(energy)*2,255), -100, 155);
|
|
|
|
sce->band_type[w*16+g] = NOISE_BT;
|
|
|
|
nz= 1;
|
|
|
|
} else { /** Band type will be determined by the twoloop algorithm */
|
|
|
|
sce->band_type[w*16+g] = 0;
|
|
|
|
}
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->zeroes[w*16+g] = !nz;
|
2009-07-08 23:36:45 +03:00
|
|
|
if (nz)
|
2009-07-09 20:22:59 +03:00
|
|
|
minthr = FFMIN(minthr, uplim);
|
2011-03-11 04:39:24 +02:00
|
|
|
allz |= nz;
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
start += sce->ics.swb_sizes[g];
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
}
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
|
|
|
|
for (g = 0; g < sce->ics.num_swb; g++) {
|
|
|
|
if (sce->zeroes[w*16+g]) {
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->sf_idx[w*16+g] = SCALE_ONE_POS;
|
|
|
|
continue;
|
|
|
|
}
|
2010-07-24 05:10:59 +03:00
|
|
|
sce->sf_idx[w*16+g] = SCALE_ONE_POS + FFMIN(log2f(uplims[w*16+g]/minthr)*4,59);
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2009-07-08 23:36:45 +03:00
|
|
|
if (!allz)
|
2009-07-08 23:01:31 +03:00
|
|
|
return;
|
|
|
|
abs_pow34_v(s->scoefs, sce->coeffs, 1024);
|
2010-07-24 20:11:51 +03:00
|
|
|
|
|
|
|
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
|
|
|
|
start = w*128;
|
|
|
|
for (g = 0; g < sce->ics.num_swb; g++) {
|
|
|
|
const float *scaled = s->scoefs + start;
|
|
|
|
maxvals[w*16+g] = find_max_val(sce->ics.group_len[w], sce->ics.swb_sizes[g], scaled);
|
|
|
|
start += sce->ics.swb_sizes[g];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2009-07-08 23:01:31 +03:00
|
|
|
//perform two-loop search
|
|
|
|
//outer loop - improve quality
|
2009-07-09 00:16:06 +03:00
|
|
|
do {
|
2009-07-08 23:01:31 +03:00
|
|
|
int tbits, qstep;
|
|
|
|
minscaler = sce->sf_idx[0];
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
minscaler_n = sce->sf_idx[0];
|
2009-07-08 23:01:31 +03:00
|
|
|
//inner loop - quantize spectrum to fit into given number of bits
|
|
|
|
qstep = its ? 1 : 32;
|
2009-07-09 00:16:06 +03:00
|
|
|
do {
|
2009-07-08 23:01:31 +03:00
|
|
|
int prev = -1;
|
|
|
|
tbits = 0;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
|
2009-07-08 23:01:31 +03:00
|
|
|
start = w*128;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (g = 0; g < sce->ics.num_swb; g++) {
|
2009-07-08 23:01:31 +03:00
|
|
|
const float *coefs = sce->coeffs + start;
|
|
|
|
const float *scaled = s->scoefs + start;
|
|
|
|
int bits = 0;
|
|
|
|
int cb;
|
2010-05-28 23:42:01 +03:00
|
|
|
float dist = 0.0f;
|
2009-07-08 23:01:31 +03:00
|
|
|
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
if (sce->band_type[w*16+g] == NOISE_BT) {
|
|
|
|
minscaler_n = FFMIN(minscaler_n, noise_sf[w*16+g]);
|
|
|
|
start += sce->ics.swb_sizes[g];
|
|
|
|
continue;
|
|
|
|
} else if (sce->zeroes[w*16+g] || sce->sf_idx[w*16+g] >= 218) {
|
2009-07-22 06:53:30 +03:00
|
|
|
start += sce->ics.swb_sizes[g];
|
2009-07-08 23:01:31 +03:00
|
|
|
continue;
|
2009-07-22 06:53:30 +03:00
|
|
|
}
|
2009-07-08 23:01:31 +03:00
|
|
|
minscaler = FFMIN(minscaler, sce->sf_idx[w*16+g]);
|
2010-07-24 20:11:51 +03:00
|
|
|
cb = find_min_book(maxvals[w*16+g], sce->sf_idx[w*16+g]);
|
2010-05-28 23:45:35 +03:00
|
|
|
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
|
|
|
|
int b;
|
|
|
|
dist += quantize_band_cost(s, coefs + w2*128,
|
|
|
|
scaled + w2*128,
|
|
|
|
sce->ics.swb_sizes[g],
|
|
|
|
sce->sf_idx[w*16+g],
|
|
|
|
cb,
|
2010-05-29 00:29:56 +03:00
|
|
|
1.0f,
|
2010-05-28 23:45:35 +03:00
|
|
|
INFINITY,
|
|
|
|
&b);
|
|
|
|
bits += b;
|
|
|
|
}
|
2010-05-29 00:29:56 +03:00
|
|
|
dists[w*16+g] = dist - bits;
|
2009-07-08 23:36:45 +03:00
|
|
|
if (prev != -1) {
|
2009-07-08 23:01:31 +03:00
|
|
|
bits += ff_aac_scalefactor_bits[sce->sf_idx[w*16+g] - prev + SCALE_DIFF_ZERO];
|
|
|
|
}
|
|
|
|
tbits += bits;
|
|
|
|
start += sce->ics.swb_sizes[g];
|
|
|
|
prev = sce->sf_idx[w*16+g];
|
|
|
|
}
|
|
|
|
}
|
2009-07-08 23:36:45 +03:00
|
|
|
if (tbits > destbits) {
|
2009-07-09 00:36:33 +03:00
|
|
|
for (i = 0; i < 128; i++)
|
|
|
|
if (sce->sf_idx[i] < 218 - qstep)
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->sf_idx[i] += qstep;
|
2009-07-08 23:36:45 +03:00
|
|
|
} else {
|
2009-07-09 00:36:33 +03:00
|
|
|
for (i = 0; i < 128; i++)
|
|
|
|
if (sce->sf_idx[i] > 60 - qstep)
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->sf_idx[i] -= qstep;
|
|
|
|
}
|
|
|
|
qstep >>= 1;
|
2010-07-24 07:23:26 +03:00
|
|
|
if (!qstep && tbits > destbits*1.02 && sce->sf_idx[0] < 217)
|
2009-07-08 23:01:31 +03:00
|
|
|
qstep = 1;
|
2009-07-09 00:16:06 +03:00
|
|
|
} while (qstep);
|
2009-07-08 23:01:31 +03:00
|
|
|
|
|
|
|
fflag = 0;
|
|
|
|
minscaler = av_clip(minscaler, 60, 255 - SCALE_MAX_DIFF);
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
|
|
|
|
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
|
|
|
|
for (g = 0; g < sce->ics.num_swb; g++)
|
|
|
|
if (sce->band_type[w*16+g] == NOISE_BT)
|
|
|
|
sce->sf_idx[w*16+g] = av_clip(noise_sf[w*16+g], minscaler_n, minscaler_n + SCALE_MAX_DIFF);
|
|
|
|
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
|
|
|
|
for (g = 0; g < sce->ics.num_swb; g++) {
|
2009-07-08 23:01:31 +03:00
|
|
|
int prevsc = sce->sf_idx[w*16+g];
|
aaccoder: Implement Perceptual Noise Substitution for AAC
This commit implements the perceptual noise substitution AAC extension. This is a proof of concept
implementation, and as such, is not enabled by default. This is the fourth revision of this patch,
made after some problems were noted out. Any changes made since the previous revisions have been indicated.
In order to extend the encoder to use an additional codebook, the array holding each codebook has been
modified with two additional entries - 13 for the NOISE_BT codebook and 12 which has a placeholder function.
The cost system was modified to skip the 12th entry using an array to map the input and outputs it has. It
also does not accept using the 13th codebook for any band which is not marked as containing noise, thereby
restricting its ability to arbitrarily choose it for bands. The use of arrays allows the system to be easily
extended to allow for intensity stereo encoding, which uses additional codebooks.
The 12th entry in the codebook function array points to a function which stops the execution of the program
by calling an assert with an always 'false' argument. It was pointed out in an email discussion with
Claudio Freire that having a 'NULL' entry can result in unexpected behaviour and could be used as
a security hole. There is no danger of this function being called during encoding due to the codebook maps introduced.
Another change from version 1 of the patch is the addition of an argument to the encoder, '-aac_pns' to
enable and disable the PNS. This currently defaults to disable the PNS, as it is experimental.
The switch will be removed in the future, when the algorithm to select noise bands has been improved.
The current algorithm simply compares the energy to the threshold (multiplied by a constant) to determine
noise, however the FFPsyBand structure contains other useful figures to determine which bands carry noise more accurately.
Some of the sample files provided triggered an assertion when the parameter to tune the threshold was set to
a value of '2.2'. Claudio Freire reported the problem's source could be in the range of the scalefactor
indices for noise and advised to measure the minimal index and clip anything above the maximum allowed
value. This has been implemented and all the files which used to trigger the asserion now encode without error.
The third revision of the problem also removes unneded variabes and comparisons. All of them were
redundant and were of little use for when the PNS implementation would be extended.
The fourth revision moved the clipping of the noise scalefactors outside the second loop of the two-loop
algorithm in order to prevent their redundant calculations. Also, freq_mult has been changed to a float
variable due to the fact that rounding errors can prove to be a problem at low frequencies.
Considerations were taken whether the entire expression could be evaluated inside the expression
, but in the end it was decided that it would be for the best if just the type of the variable were
to change. Claudio Freire reported the two problems. There is no change of functionality
(except for low sampling frequencies) so the spectral demonstrations at the end of this commit's message were not updated.
Finally, the way energy values are converted to scalefactor indices has changed since the first commit,
as per the suggestion of Claudio Freire. This may still have some drawbacks, but unlike the first commit
it works without having redundant offsets and outputs what the decoder expects to have, in terms of the
ranges of the scalefactor indices.
Some spectral comparisons: https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Original.png (original),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_NO.png (encoded without PNS),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS1.2.png (encoded with PNS, const = 1.2),
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/Difference1.png (spectral difference).
The constant is the value which multiplies the threshold when it gets compared to the energy, larger
values means more noise will be substituded by PNS values. Example when const = 2.2:
https://trac.ffmpeg.org/attachment/wiki/Encode/AAC/PNS_2.2.png
Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-04-15 13:18:42 +02:00
|
|
|
if (sce->band_type[w*16+g] == NOISE_BT)
|
|
|
|
continue;
|
2010-07-24 19:36:01 +03:00
|
|
|
if (dists[w*16+g] > uplims[w*16+g] && sce->sf_idx[w*16+g] > 60) {
|
2010-07-24 20:11:51 +03:00
|
|
|
if (find_min_book(maxvals[w*16+g], sce->sf_idx[w*16+g]-1))
|
2011-03-25 05:17:48 +02:00
|
|
|
sce->sf_idx[w*16+g]--;
|
2010-07-24 19:36:01 +03:00
|
|
|
else //Try to make sure there is some energy in every band
|
|
|
|
sce->sf_idx[w*16+g]-=2;
|
|
|
|
}
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->sf_idx[w*16+g] = av_clip(sce->sf_idx[w*16+g], minscaler, minscaler + SCALE_MAX_DIFF);
|
|
|
|
sce->sf_idx[w*16+g] = FFMIN(sce->sf_idx[w*16+g], 219);
|
2009-07-08 23:36:45 +03:00
|
|
|
if (sce->sf_idx[w*16+g] != prevsc)
|
2009-07-08 23:01:31 +03:00
|
|
|
fflag = 1;
|
2010-07-24 20:11:51 +03:00
|
|
|
sce->band_type[w*16+g] = find_min_book(maxvals[w*16+g], sce->sf_idx[w*16+g]);
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
its++;
|
2009-07-09 00:16:06 +03:00
|
|
|
} while (fflag && its < 10);
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
static void search_for_quantizers_faac(AVCodecContext *avctx, AACEncContext *s,
|
2009-07-09 00:16:06 +03:00
|
|
|
SingleChannelElement *sce,
|
|
|
|
const float lambda)
|
2009-07-08 23:01:31 +03:00
|
|
|
{
|
|
|
|
int start = 0, i, w, w2, g;
|
|
|
|
float uplim[128], maxq[128];
|
|
|
|
int minq, maxsf;
|
|
|
|
float distfact = ((sce->ics.num_windows > 1) ? 85.80 : 147.84) / lambda;
|
|
|
|
int last = 0, lastband = 0, curband = 0;
|
|
|
|
float avg_energy = 0.0;
|
2009-07-08 23:36:45 +03:00
|
|
|
if (sce->ics.num_windows == 1) {
|
2009-07-08 23:01:31 +03:00
|
|
|
start = 0;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (i = 0; i < 1024; i++) {
|
|
|
|
if (i - start >= sce->ics.swb_sizes[curband]) {
|
2009-07-08 23:01:31 +03:00
|
|
|
start += sce->ics.swb_sizes[curband];
|
|
|
|
curband++;
|
|
|
|
}
|
2009-07-08 23:36:45 +03:00
|
|
|
if (sce->coeffs[i]) {
|
2009-07-08 23:01:31 +03:00
|
|
|
avg_energy += sce->coeffs[i] * sce->coeffs[i];
|
|
|
|
last = i;
|
|
|
|
lastband = curband;
|
|
|
|
}
|
|
|
|
}
|
2009-07-08 23:36:45 +03:00
|
|
|
} else {
|
|
|
|
for (w = 0; w < 8; w++) {
|
2009-07-08 23:01:31 +03:00
|
|
|
const float *coeffs = sce->coeffs + w*128;
|
2012-06-06 19:35:05 +03:00
|
|
|
curband = start = 0;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (i = 0; i < 128; i++) {
|
|
|
|
if (i - start >= sce->ics.swb_sizes[curband]) {
|
2009-07-08 23:01:31 +03:00
|
|
|
start += sce->ics.swb_sizes[curband];
|
|
|
|
curband++;
|
|
|
|
}
|
2009-07-08 23:36:45 +03:00
|
|
|
if (coeffs[i]) {
|
2009-07-08 23:01:31 +03:00
|
|
|
avg_energy += coeffs[i] * coeffs[i];
|
|
|
|
last = FFMAX(last, i);
|
|
|
|
lastband = FFMAX(lastband, curband);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
last++;
|
|
|
|
avg_energy /= last;
|
2009-07-08 23:36:45 +03:00
|
|
|
if (avg_energy == 0.0f) {
|
|
|
|
for (i = 0; i < FF_ARRAY_ELEMS(sce->sf_idx); i++)
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->sf_idx[i] = SCALE_ONE_POS;
|
|
|
|
return;
|
|
|
|
}
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
|
2009-07-08 23:01:31 +03:00
|
|
|
start = w*128;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (g = 0; g < sce->ics.num_swb; g++) {
|
2009-07-09 00:16:06 +03:00
|
|
|
float *coefs = sce->coeffs + start;
|
2009-07-08 23:01:31 +03:00
|
|
|
const int size = sce->ics.swb_sizes[g];
|
|
|
|
int start2 = start, end2 = start + size, peakpos = start;
|
|
|
|
float maxval = -1, thr = 0.0f, t;
|
|
|
|
maxq[w*16+g] = 0.0f;
|
2009-07-08 23:36:45 +03:00
|
|
|
if (g > lastband) {
|
2009-07-08 23:01:31 +03:00
|
|
|
maxq[w*16+g] = 0.0f;
|
|
|
|
start += size;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++)
|
2009-07-08 23:01:31 +03:00
|
|
|
memset(coefs + w2*128, 0, sizeof(coefs[0])*size);
|
|
|
|
continue;
|
|
|
|
}
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
|
|
|
|
for (i = 0; i < size; i++) {
|
2009-07-08 23:01:31 +03:00
|
|
|
float t = coefs[w2*128+i]*coefs[w2*128+i];
|
2009-07-09 20:22:59 +03:00
|
|
|
maxq[w*16+g] = FFMAX(maxq[w*16+g], fabsf(coefs[w2*128 + i]));
|
2009-07-08 23:01:31 +03:00
|
|
|
thr += t;
|
2009-07-08 23:36:45 +03:00
|
|
|
if (sce->ics.num_windows == 1 && maxval < t) {
|
2009-07-09 00:16:06 +03:00
|
|
|
maxval = t;
|
2009-07-08 23:01:31 +03:00
|
|
|
peakpos = start+i;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2009-07-08 23:36:45 +03:00
|
|
|
if (sce->ics.num_windows == 1) {
|
2009-07-08 23:01:31 +03:00
|
|
|
start2 = FFMAX(peakpos - 2, start2);
|
|
|
|
end2 = FFMIN(peakpos + 3, end2);
|
2009-07-08 23:36:45 +03:00
|
|
|
} else {
|
2009-07-08 23:01:31 +03:00
|
|
|
start2 -= start;
|
|
|
|
end2 -= start;
|
|
|
|
}
|
|
|
|
start += size;
|
|
|
|
thr = pow(thr / (avg_energy * (end2 - start2)), 0.3 + 0.1*(lastband - g) / lastband);
|
2009-07-09 00:16:06 +03:00
|
|
|
t = 1.0 - (1.0 * start2 / last);
|
2009-07-08 23:01:31 +03:00
|
|
|
uplim[w*16+g] = distfact / (1.4 * thr + t*t*t + 0.075);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
memset(sce->sf_idx, 0, sizeof(sce->sf_idx));
|
|
|
|
abs_pow34_v(s->scoefs, sce->coeffs, 1024);
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
|
2009-07-08 23:01:31 +03:00
|
|
|
start = w*128;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (g = 0; g < sce->ics.num_swb; g++) {
|
2009-07-09 00:16:06 +03:00
|
|
|
const float *coefs = sce->coeffs + start;
|
|
|
|
const float *scaled = s->scoefs + start;
|
|
|
|
const int size = sce->ics.swb_sizes[g];
|
2009-07-08 23:01:31 +03:00
|
|
|
int scf, prev_scf, step;
|
2010-05-06 23:18:36 +03:00
|
|
|
int min_scf = -1, max_scf = 256;
|
2009-07-08 23:01:31 +03:00
|
|
|
float curdiff;
|
2009-07-08 23:36:45 +03:00
|
|
|
if (maxq[w*16+g] < 21.544) {
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->zeroes[w*16+g] = 1;
|
|
|
|
start += size;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
sce->zeroes[w*16+g] = 0;
|
2010-07-24 05:10:59 +03:00
|
|
|
scf = prev_scf = av_clip(SCALE_ONE_POS - SCALE_DIV_512 - log2f(1/maxq[w*16+g])*16/3, 60, 218);
|
2009-07-08 23:36:45 +03:00
|
|
|
for (;;) {
|
2009-07-08 23:01:31 +03:00
|
|
|
float dist = 0.0f;
|
|
|
|
int quant_max;
|
|
|
|
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
|
2009-07-08 23:01:31 +03:00
|
|
|
int b;
|
|
|
|
dist += quantize_band_cost(s, coefs + w2*128,
|
|
|
|
scaled + w2*128,
|
|
|
|
sce->ics.swb_sizes[g],
|
|
|
|
scf,
|
|
|
|
ESC_BT,
|
2009-07-09 02:04:22 +03:00
|
|
|
lambda,
|
2009-07-08 23:01:31 +03:00
|
|
|
INFINITY,
|
|
|
|
&b);
|
|
|
|
dist -= b;
|
|
|
|
}
|
2009-07-09 02:04:22 +03:00
|
|
|
dist *= 1.0f / 512.0f / lambda;
|
2011-04-26 21:45:48 +03:00
|
|
|
quant_max = quant(maxq[w*16+g], ff_aac_pow2sf_tab[POW_SF2_ZERO - scf + SCALE_ONE_POS - SCALE_DIV_512]);
|
2009-07-08 23:36:45 +03:00
|
|
|
if (quant_max >= 8191) { // too much, return to the previous quantizer
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->sf_idx[w*16+g] = prev_scf;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
prev_scf = scf;
|
|
|
|
curdiff = fabsf(dist - uplim[w*16+g]);
|
2010-05-06 23:18:36 +03:00
|
|
|
if (curdiff <= 1.0f)
|
2009-07-08 23:01:31 +03:00
|
|
|
step = 0;
|
|
|
|
else
|
2010-07-24 05:10:59 +03:00
|
|
|
step = log2f(curdiff);
|
2009-07-08 23:36:45 +03:00
|
|
|
if (dist > uplim[w*16+g])
|
2009-07-08 23:01:31 +03:00
|
|
|
step = -step;
|
2010-05-06 23:18:36 +03:00
|
|
|
scf += step;
|
2010-05-06 23:38:42 +03:00
|
|
|
scf = av_clip_uint8(scf);
|
2010-05-06 23:18:36 +03:00
|
|
|
step = scf - prev_scf;
|
2009-07-08 23:36:45 +03:00
|
|
|
if (FFABS(step) <= 1 || (step > 0 && scf >= max_scf) || (step < 0 && scf <= min_scf)) {
|
2010-05-06 23:18:36 +03:00
|
|
|
sce->sf_idx[w*16+g] = av_clip(scf, min_scf, max_scf);
|
2009-07-08 23:01:31 +03:00
|
|
|
break;
|
|
|
|
}
|
2009-07-08 23:36:45 +03:00
|
|
|
if (step > 0)
|
2010-05-06 23:18:36 +03:00
|
|
|
min_scf = prev_scf;
|
2009-07-08 23:01:31 +03:00
|
|
|
else
|
2010-05-06 23:18:36 +03:00
|
|
|
max_scf = prev_scf;
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
start += size;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
minq = sce->sf_idx[0] ? sce->sf_idx[0] : INT_MAX;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (i = 1; i < 128; i++) {
|
|
|
|
if (!sce->sf_idx[i])
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->sf_idx[i] = sce->sf_idx[i-1];
|
|
|
|
else
|
|
|
|
minq = FFMIN(minq, sce->sf_idx[i]);
|
|
|
|
}
|
2009-07-09 00:36:33 +03:00
|
|
|
if (minq == INT_MAX)
|
|
|
|
minq = 0;
|
2009-07-08 23:01:31 +03:00
|
|
|
minq = FFMIN(minq, SCALE_MAX_POS);
|
|
|
|
maxsf = FFMIN(minq + SCALE_MAX_DIFF, SCALE_MAX_POS);
|
2009-07-08 23:36:45 +03:00
|
|
|
for (i = 126; i >= 0; i--) {
|
|
|
|
if (!sce->sf_idx[i])
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->sf_idx[i] = sce->sf_idx[i+1];
|
|
|
|
sce->sf_idx[i] = av_clip(sce->sf_idx[i], minq, maxsf);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void search_for_quantizers_fast(AVCodecContext *avctx, AACEncContext *s,
|
2009-07-09 00:16:06 +03:00
|
|
|
SingleChannelElement *sce,
|
|
|
|
const float lambda)
|
2009-07-08 23:01:31 +03:00
|
|
|
{
|
2011-06-01 19:26:27 +03:00
|
|
|
int i, w, w2, g;
|
2009-07-08 23:01:31 +03:00
|
|
|
int minq = 255;
|
|
|
|
|
|
|
|
memset(sce->sf_idx, 0, sizeof(sce->sf_idx));
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
|
|
|
|
for (g = 0; g < sce->ics.num_swb; g++) {
|
|
|
|
for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {
|
2011-06-15 11:50:25 +03:00
|
|
|
FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];
|
2009-07-08 23:36:45 +03:00
|
|
|
if (band->energy <= band->threshold) {
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->sf_idx[(w+w2)*16+g] = 218;
|
|
|
|
sce->zeroes[(w+w2)*16+g] = 1;
|
2009-07-08 23:36:45 +03:00
|
|
|
} else {
|
2010-07-24 05:10:59 +03:00
|
|
|
sce->sf_idx[(w+w2)*16+g] = av_clip(SCALE_ONE_POS - SCALE_DIV_512 + log2f(band->threshold), 80, 218);
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->zeroes[(w+w2)*16+g] = 0;
|
|
|
|
}
|
|
|
|
minq = FFMIN(minq, sce->sf_idx[(w+w2)*16+g]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2009-07-08 23:36:45 +03:00
|
|
|
for (i = 0; i < 128; i++) {
|
2009-07-09 00:36:33 +03:00
|
|
|
sce->sf_idx[i] = 140;
|
|
|
|
//av_clip(sce->sf_idx[i], minq, minq + SCALE_MAX_DIFF - 1);
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
//set the same quantizers inside window groups
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
|
|
|
|
for (g = 0; g < sce->ics.num_swb; g++)
|
|
|
|
for (w2 = 1; w2 < sce->ics.group_len[w]; w2++)
|
2009-07-08 23:01:31 +03:00
|
|
|
sce->sf_idx[(w+w2)*16+g] = sce->sf_idx[w*16+g];
|
|
|
|
}
|
|
|
|
|
2009-07-09 00:16:06 +03:00
|
|
|
static void search_for_ms(AACEncContext *s, ChannelElement *cpe,
|
|
|
|
const float lambda)
|
2009-07-08 23:01:31 +03:00
|
|
|
{
|
|
|
|
int start = 0, i, w, w2, g;
|
|
|
|
float M[128], S[128];
|
|
|
|
float *L34 = s->scoefs, *R34 = s->scoefs + 128, *M34 = s->scoefs + 128*2, *S34 = s->scoefs + 128*3;
|
|
|
|
SingleChannelElement *sce0 = &cpe->ch[0];
|
|
|
|
SingleChannelElement *sce1 = &cpe->ch[1];
|
2009-07-08 23:36:45 +03:00
|
|
|
if (!cpe->common_window)
|
2009-07-08 23:01:31 +03:00
|
|
|
return;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w = 0; w < sce0->ics.num_windows; w += sce0->ics.group_len[w]) {
|
|
|
|
for (g = 0; g < sce0->ics.num_swb; g++) {
|
|
|
|
if (!cpe->ch[0].zeroes[w*16+g] && !cpe->ch[1].zeroes[w*16+g]) {
|
2009-07-08 23:01:31 +03:00
|
|
|
float dist1 = 0.0f, dist2 = 0.0f;
|
2009-07-08 23:36:45 +03:00
|
|
|
for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
|
2011-06-15 11:50:25 +03:00
|
|
|
FFPsyBand *band0 = &s->psy.ch[s->cur_channel+0].psy_bands[(w+w2)*16+g];
|
|
|
|
FFPsyBand *band1 = &s->psy.ch[s->cur_channel+1].psy_bands[(w+w2)*16+g];
|
2009-07-09 20:22:59 +03:00
|
|
|
float minthr = FFMIN(band0->threshold, band1->threshold);
|
|
|
|
float maxthr = FFMAX(band0->threshold, band1->threshold);
|
2009-07-08 23:36:45 +03:00
|
|
|
for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
|
2015-03-03 08:43:06 +02:00
|
|
|
M[i] = (sce0->pcoeffs[start+w2*128+i]
|
|
|
|
+ sce1->pcoeffs[start+w2*128+i]) * 0.5;
|
2011-03-24 03:49:36 +02:00
|
|
|
S[i] = M[i]
|
2015-03-03 08:43:06 +02:00
|
|
|
- sce1->pcoeffs[start+w2*128+i];
|
2009-07-08 23:01:31 +03:00
|
|
|
}
|
|
|
|
abs_pow34_v(L34, sce0->coeffs+start+w2*128, sce0->ics.swb_sizes[g]);
|
|
|
|
abs_pow34_v(R34, sce1->coeffs+start+w2*128, sce0->ics.swb_sizes[g]);
|
|
|
|
abs_pow34_v(M34, M, sce0->ics.swb_sizes[g]);
|
|
|
|
abs_pow34_v(S34, S, sce0->ics.swb_sizes[g]);
|
|
|
|
dist1 += quantize_band_cost(s, sce0->coeffs + start + w2*128,
|
|
|
|
L34,
|
|
|
|
sce0->ics.swb_sizes[g],
|
|
|
|
sce0->sf_idx[(w+w2)*16+g],
|
|
|
|
sce0->band_type[(w+w2)*16+g],
|
|
|
|
lambda / band0->threshold, INFINITY, NULL);
|
|
|
|
dist1 += quantize_band_cost(s, sce1->coeffs + start + w2*128,
|
|
|
|
R34,
|
|
|
|
sce1->ics.swb_sizes[g],
|
|
|
|
sce1->sf_idx[(w+w2)*16+g],
|
|
|
|
sce1->band_type[(w+w2)*16+g],
|
|
|
|
lambda / band1->threshold, INFINITY, NULL);
|
|
|
|
dist2 += quantize_band_cost(s, M,
|
|
|
|
M34,
|
|
|
|
sce0->ics.swb_sizes[g],
|
|
|
|
sce0->sf_idx[(w+w2)*16+g],
|
|
|
|
sce0->band_type[(w+w2)*16+g],
|
|
|
|
lambda / maxthr, INFINITY, NULL);
|
|
|
|
dist2 += quantize_band_cost(s, S,
|
|
|
|
S34,
|
|
|
|
sce1->ics.swb_sizes[g],
|
|
|
|
sce1->sf_idx[(w+w2)*16+g],
|
|
|
|
sce1->band_type[(w+w2)*16+g],
|
|
|
|
lambda / minthr, INFINITY, NULL);
|
|
|
|
}
|
|
|
|
cpe->ms_mask[w*16+g] = dist2 < dist1;
|
|
|
|
}
|
|
|
|
start += sce0->ics.swb_sizes[g];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2011-11-27 23:19:07 +03:00
|
|
|
AACCoefficientsEncoder ff_aac_coders[AAC_CODER_NB] = {
|
2013-09-11 06:23:32 +03:00
|
|
|
[AAC_CODER_FAAC] = {
|
2009-07-08 23:01:31 +03:00
|
|
|
search_for_quantizers_faac,
|
2009-07-09 02:04:22 +03:00
|
|
|
encode_window_bands_info,
|
2009-07-08 23:01:31 +03:00
|
|
|
quantize_and_encode_band,
|
2009-08-27 20:46:13 +03:00
|
|
|
search_for_ms,
|
2009-07-08 23:01:31 +03:00
|
|
|
},
|
2013-09-11 06:23:32 +03:00
|
|
|
[AAC_CODER_ANMR] = {
|
2009-07-08 23:01:31 +03:00
|
|
|
search_for_quantizers_anmr,
|
|
|
|
encode_window_bands_info,
|
|
|
|
quantize_and_encode_band,
|
2009-08-27 20:46:13 +03:00
|
|
|
search_for_ms,
|
2009-07-08 23:01:31 +03:00
|
|
|
},
|
2013-09-11 06:23:32 +03:00
|
|
|
[AAC_CODER_TWOLOOP] = {
|
2009-07-08 23:01:31 +03:00
|
|
|
search_for_quantizers_twoloop,
|
2010-05-14 19:49:51 +03:00
|
|
|
codebook_trellis_rate,
|
2009-07-08 23:01:31 +03:00
|
|
|
quantize_and_encode_band,
|
2009-08-27 20:46:13 +03:00
|
|
|
search_for_ms,
|
2009-07-08 23:01:31 +03:00
|
|
|
},
|
2013-09-11 06:23:32 +03:00
|
|
|
[AAC_CODER_FAST] = {
|
2009-07-08 23:01:31 +03:00
|
|
|
search_for_quantizers_fast,
|
|
|
|
encode_window_bands_info,
|
|
|
|
quantize_and_encode_band,
|
2009-08-27 20:46:13 +03:00
|
|
|
search_for_ms,
|
2009-07-08 23:01:31 +03:00
|
|
|
},
|
|
|
|
};
|