mirror of
https://github.com/FFmpeg/FFmpeg.git
synced 2024-11-26 19:01:44 +02:00
d5202e4fda
patch by Stefano Sabatini, stefano.sabatini-lala poste it Originally committed as revision 13005 to svn://svn.ffmpeg.org/ffmpeg/trunk
880 lines
27 KiB
C
880 lines
27 KiB
C
/*
|
|
* WMA compatible decoder
|
|
* Copyright (c) 2002 The FFmpeg Project.
|
|
*
|
|
* 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
|
|
*/
|
|
|
|
/**
|
|
* @file wmadec.c
|
|
* WMA compatible decoder.
|
|
* This decoder handles Microsoft Windows Media Audio data, versions 1 & 2.
|
|
* WMA v1 is identified by audio format 0x160 in Microsoft media files
|
|
* (ASF/AVI/WAV). WMA v2 is identified by audio format 0x161.
|
|
*
|
|
* To use this decoder, a calling application must supply the extra data
|
|
* bytes provided with the WMA data. These are the extra, codec-specific
|
|
* bytes at the end of a WAVEFORMATEX data structure. Transmit these bytes
|
|
* to the decoder using the extradata[_size] fields in AVCodecContext. There
|
|
* should be 4 extra bytes for v1 data and 6 extra bytes for v2 data.
|
|
*/
|
|
|
|
#include "avcodec.h"
|
|
#include "wma.h"
|
|
|
|
#undef NDEBUG
|
|
#include <assert.h>
|
|
|
|
#define EXPVLCBITS 8
|
|
#define EXPMAX ((19+EXPVLCBITS-1)/EXPVLCBITS)
|
|
|
|
#define HGAINVLCBITS 9
|
|
#define HGAINMAX ((13+HGAINVLCBITS-1)/HGAINVLCBITS)
|
|
|
|
static void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len);
|
|
|
|
#ifdef TRACE
|
|
static void dump_shorts(WMACodecContext *s, const char *name, const short *tab, int n)
|
|
{
|
|
int i;
|
|
|
|
tprintf(s->avctx, "%s[%d]:\n", name, n);
|
|
for(i=0;i<n;i++) {
|
|
if ((i & 7) == 0)
|
|
tprintf(s->avctx, "%4d: ", i);
|
|
tprintf(s->avctx, " %5d.0", tab[i]);
|
|
if ((i & 7) == 7)
|
|
tprintf(s->avctx, "\n");
|
|
}
|
|
}
|
|
|
|
static void dump_floats(WMACodecContext *s, const char *name, int prec, const float *tab, int n)
|
|
{
|
|
int i;
|
|
|
|
tprintf(s->avctx, "%s[%d]:\n", name, n);
|
|
for(i=0;i<n;i++) {
|
|
if ((i & 7) == 0)
|
|
tprintf(s->avctx, "%4d: ", i);
|
|
tprintf(s->avctx, " %8.*f", prec, tab[i]);
|
|
if ((i & 7) == 7)
|
|
tprintf(s->avctx, "\n");
|
|
}
|
|
if ((i & 7) != 0)
|
|
tprintf(s->avctx, "\n");
|
|
}
|
|
#endif
|
|
|
|
static int wma_decode_init(AVCodecContext * avctx)
|
|
{
|
|
WMACodecContext *s = avctx->priv_data;
|
|
int i, flags1, flags2;
|
|
uint8_t *extradata;
|
|
|
|
s->avctx = avctx;
|
|
|
|
/* extract flag infos */
|
|
flags1 = 0;
|
|
flags2 = 0;
|
|
extradata = avctx->extradata;
|
|
if (avctx->codec->id == CODEC_ID_WMAV1 && avctx->extradata_size >= 4) {
|
|
flags1 = AV_RL16(extradata);
|
|
flags2 = AV_RL16(extradata+2);
|
|
} else if (avctx->codec->id == CODEC_ID_WMAV2 && avctx->extradata_size >= 6) {
|
|
flags1 = AV_RL32(extradata);
|
|
flags2 = AV_RL16(extradata+4);
|
|
}
|
|
// for(i=0; i<avctx->extradata_size; i++)
|
|
// av_log(NULL, AV_LOG_ERROR, "%02X ", extradata[i]);
|
|
|
|
s->use_exp_vlc = flags2 & 0x0001;
|
|
s->use_bit_reservoir = flags2 & 0x0002;
|
|
s->use_variable_block_len = flags2 & 0x0004;
|
|
|
|
if(ff_wma_init(avctx, flags2)<0)
|
|
return -1;
|
|
|
|
/* init MDCT */
|
|
for(i = 0; i < s->nb_block_sizes; i++)
|
|
ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 1);
|
|
|
|
if (s->use_noise_coding) {
|
|
init_vlc(&s->hgain_vlc, HGAINVLCBITS, sizeof(ff_wma_hgain_huffbits),
|
|
ff_wma_hgain_huffbits, 1, 1,
|
|
ff_wma_hgain_huffcodes, 2, 2, 0);
|
|
}
|
|
|
|
if (s->use_exp_vlc) {
|
|
init_vlc(&s->exp_vlc, EXPVLCBITS, sizeof(ff_wma_scale_huffbits), //FIXME move out of context
|
|
ff_wma_scale_huffbits, 1, 1,
|
|
ff_wma_scale_huffcodes, 4, 4, 0);
|
|
} else {
|
|
wma_lsp_to_curve_init(s, s->frame_len);
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
/**
|
|
* compute x^-0.25 with an exponent and mantissa table. We use linear
|
|
* interpolation to reduce the mantissa table size at a small speed
|
|
* expense (linear interpolation approximately doubles the number of
|
|
* bits of precision).
|
|
*/
|
|
static inline float pow_m1_4(WMACodecContext *s, float x)
|
|
{
|
|
union {
|
|
float f;
|
|
unsigned int v;
|
|
} u, t;
|
|
unsigned int e, m;
|
|
float a, b;
|
|
|
|
u.f = x;
|
|
e = u.v >> 23;
|
|
m = (u.v >> (23 - LSP_POW_BITS)) & ((1 << LSP_POW_BITS) - 1);
|
|
/* build interpolation scale: 1 <= t < 2. */
|
|
t.v = ((u.v << LSP_POW_BITS) & ((1 << 23) - 1)) | (127 << 23);
|
|
a = s->lsp_pow_m_table1[m];
|
|
b = s->lsp_pow_m_table2[m];
|
|
return s->lsp_pow_e_table[e] * (a + b * t.f);
|
|
}
|
|
|
|
static void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len)
|
|
{
|
|
float wdel, a, b;
|
|
int i, e, m;
|
|
|
|
wdel = M_PI / frame_len;
|
|
for(i=0;i<frame_len;i++)
|
|
s->lsp_cos_table[i] = 2.0f * cos(wdel * i);
|
|
|
|
/* tables for x^-0.25 computation */
|
|
for(i=0;i<256;i++) {
|
|
e = i - 126;
|
|
s->lsp_pow_e_table[i] = pow(2.0, e * -0.25);
|
|
}
|
|
|
|
/* NOTE: these two tables are needed to avoid two operations in
|
|
pow_m1_4 */
|
|
b = 1.0;
|
|
for(i=(1 << LSP_POW_BITS) - 1;i>=0;i--) {
|
|
m = (1 << LSP_POW_BITS) + i;
|
|
a = (float)m * (0.5 / (1 << LSP_POW_BITS));
|
|
a = pow(a, -0.25);
|
|
s->lsp_pow_m_table1[i] = 2 * a - b;
|
|
s->lsp_pow_m_table2[i] = b - a;
|
|
b = a;
|
|
}
|
|
#if 0
|
|
for(i=1;i<20;i++) {
|
|
float v, r1, r2;
|
|
v = 5.0 / i;
|
|
r1 = pow_m1_4(s, v);
|
|
r2 = pow(v,-0.25);
|
|
printf("%f^-0.25=%f e=%f\n", v, r1, r2 - r1);
|
|
}
|
|
#endif
|
|
}
|
|
|
|
/**
|
|
* NOTE: We use the same code as Vorbis here
|
|
* @todo optimize it further with SSE/3Dnow
|
|
*/
|
|
static void wma_lsp_to_curve(WMACodecContext *s,
|
|
float *out, float *val_max_ptr,
|
|
int n, float *lsp)
|
|
{
|
|
int i, j;
|
|
float p, q, w, v, val_max;
|
|
|
|
val_max = 0;
|
|
for(i=0;i<n;i++) {
|
|
p = 0.5f;
|
|
q = 0.5f;
|
|
w = s->lsp_cos_table[i];
|
|
for(j=1;j<NB_LSP_COEFS;j+=2){
|
|
q *= w - lsp[j - 1];
|
|
p *= w - lsp[j];
|
|
}
|
|
p *= p * (2.0f - w);
|
|
q *= q * (2.0f + w);
|
|
v = p + q;
|
|
v = pow_m1_4(s, v);
|
|
if (v > val_max)
|
|
val_max = v;
|
|
out[i] = v;
|
|
}
|
|
*val_max_ptr = val_max;
|
|
}
|
|
|
|
/**
|
|
* decode exponents coded with LSP coefficients (same idea as Vorbis)
|
|
*/
|
|
static void decode_exp_lsp(WMACodecContext *s, int ch)
|
|
{
|
|
float lsp_coefs[NB_LSP_COEFS];
|
|
int val, i;
|
|
|
|
for(i = 0; i < NB_LSP_COEFS; i++) {
|
|
if (i == 0 || i >= 8)
|
|
val = get_bits(&s->gb, 3);
|
|
else
|
|
val = get_bits(&s->gb, 4);
|
|
lsp_coefs[i] = ff_wma_lsp_codebook[i][val];
|
|
}
|
|
|
|
wma_lsp_to_curve(s, s->exponents[ch], &s->max_exponent[ch],
|
|
s->block_len, lsp_coefs);
|
|
}
|
|
|
|
/**
|
|
* decode exponents coded with VLC codes
|
|
*/
|
|
static int decode_exp_vlc(WMACodecContext *s, int ch)
|
|
{
|
|
int last_exp, n, code;
|
|
const uint16_t *ptr, *band_ptr;
|
|
float v, *q, max_scale, *q_end;
|
|
|
|
band_ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
|
|
ptr = band_ptr;
|
|
q = s->exponents[ch];
|
|
q_end = q + s->block_len;
|
|
max_scale = 0;
|
|
if (s->version == 1) {
|
|
last_exp = get_bits(&s->gb, 5) + 10;
|
|
/* XXX: use a table */
|
|
v = pow(10, last_exp * (1.0 / 16.0));
|
|
max_scale = v;
|
|
n = *ptr++;
|
|
do {
|
|
*q++ = v;
|
|
} while (--n);
|
|
}else
|
|
last_exp = 36;
|
|
|
|
while (q < q_end) {
|
|
code = get_vlc2(&s->gb, s->exp_vlc.table, EXPVLCBITS, EXPMAX);
|
|
if (code < 0)
|
|
return -1;
|
|
/* NOTE: this offset is the same as MPEG4 AAC ! */
|
|
last_exp += code - 60;
|
|
/* XXX: use a table */
|
|
v = pow(10, last_exp * (1.0 / 16.0));
|
|
if (v > max_scale)
|
|
max_scale = v;
|
|
n = *ptr++;
|
|
do {
|
|
*q++ = v;
|
|
} while (--n);
|
|
}
|
|
s->max_exponent[ch] = max_scale;
|
|
return 0;
|
|
}
|
|
|
|
|
|
/**
|
|
* Apply MDCT window and add into output.
|
|
*
|
|
* We ensure that when the windows overlap their squared sum
|
|
* is always 1 (MDCT reconstruction rule).
|
|
*/
|
|
static void wma_window(WMACodecContext *s, float *out)
|
|
{
|
|
float *in = s->output;
|
|
int block_len, bsize, n;
|
|
|
|
/* left part */
|
|
if (s->block_len_bits <= s->prev_block_len_bits) {
|
|
block_len = s->block_len;
|
|
bsize = s->frame_len_bits - s->block_len_bits;
|
|
|
|
s->dsp.vector_fmul_add_add(out, in, s->windows[bsize],
|
|
out, 0, block_len, 1);
|
|
|
|
} else {
|
|
block_len = 1 << s->prev_block_len_bits;
|
|
n = (s->block_len - block_len) / 2;
|
|
bsize = s->frame_len_bits - s->prev_block_len_bits;
|
|
|
|
s->dsp.vector_fmul_add_add(out+n, in+n, s->windows[bsize],
|
|
out+n, 0, block_len, 1);
|
|
|
|
memcpy(out+n+block_len, in+n+block_len, n*sizeof(float));
|
|
}
|
|
|
|
out += s->block_len;
|
|
in += s->block_len;
|
|
|
|
/* right part */
|
|
if (s->block_len_bits <= s->next_block_len_bits) {
|
|
block_len = s->block_len;
|
|
bsize = s->frame_len_bits - s->block_len_bits;
|
|
|
|
s->dsp.vector_fmul_reverse(out, in, s->windows[bsize], block_len);
|
|
|
|
} else {
|
|
block_len = 1 << s->next_block_len_bits;
|
|
n = (s->block_len - block_len) / 2;
|
|
bsize = s->frame_len_bits - s->next_block_len_bits;
|
|
|
|
memcpy(out, in, n*sizeof(float));
|
|
|
|
s->dsp.vector_fmul_reverse(out+n, in+n, s->windows[bsize], block_len);
|
|
|
|
memset(out+n+block_len, 0, n*sizeof(float));
|
|
}
|
|
}
|
|
|
|
|
|
/**
|
|
* @return 0 if OK. 1 if last block of frame. return -1 if
|
|
* unrecorrable error.
|
|
*/
|
|
static int wma_decode_block(WMACodecContext *s)
|
|
{
|
|
int n, v, a, ch, code, bsize;
|
|
int coef_nb_bits, total_gain;
|
|
int nb_coefs[MAX_CHANNELS];
|
|
float mdct_norm;
|
|
|
|
#ifdef TRACE
|
|
tprintf(s->avctx, "***decode_block: %d:%d\n", s->frame_count - 1, s->block_num);
|
|
#endif
|
|
|
|
/* compute current block length */
|
|
if (s->use_variable_block_len) {
|
|
n = av_log2(s->nb_block_sizes - 1) + 1;
|
|
|
|
if (s->reset_block_lengths) {
|
|
s->reset_block_lengths = 0;
|
|
v = get_bits(&s->gb, n);
|
|
if (v >= s->nb_block_sizes)
|
|
return -1;
|
|
s->prev_block_len_bits = s->frame_len_bits - v;
|
|
v = get_bits(&s->gb, n);
|
|
if (v >= s->nb_block_sizes)
|
|
return -1;
|
|
s->block_len_bits = s->frame_len_bits - v;
|
|
} else {
|
|
/* update block lengths */
|
|
s->prev_block_len_bits = s->block_len_bits;
|
|
s->block_len_bits = s->next_block_len_bits;
|
|
}
|
|
v = get_bits(&s->gb, n);
|
|
if (v >= s->nb_block_sizes)
|
|
return -1;
|
|
s->next_block_len_bits = s->frame_len_bits - v;
|
|
} else {
|
|
/* fixed block len */
|
|
s->next_block_len_bits = s->frame_len_bits;
|
|
s->prev_block_len_bits = s->frame_len_bits;
|
|
s->block_len_bits = s->frame_len_bits;
|
|
}
|
|
|
|
/* now check if the block length is coherent with the frame length */
|
|
s->block_len = 1 << s->block_len_bits;
|
|
if ((s->block_pos + s->block_len) > s->frame_len)
|
|
return -1;
|
|
|
|
if (s->nb_channels == 2) {
|
|
s->ms_stereo = get_bits1(&s->gb);
|
|
}
|
|
v = 0;
|
|
for(ch = 0; ch < s->nb_channels; ch++) {
|
|
a = get_bits1(&s->gb);
|
|
s->channel_coded[ch] = a;
|
|
v |= a;
|
|
}
|
|
/* if no channel coded, no need to go further */
|
|
/* XXX: fix potential framing problems */
|
|
if (!v)
|
|
goto next;
|
|
|
|
bsize = s->frame_len_bits - s->block_len_bits;
|
|
|
|
/* read total gain and extract corresponding number of bits for
|
|
coef escape coding */
|
|
total_gain = 1;
|
|
for(;;) {
|
|
a = get_bits(&s->gb, 7);
|
|
total_gain += a;
|
|
if (a != 127)
|
|
break;
|
|
}
|
|
|
|
coef_nb_bits= ff_wma_total_gain_to_bits(total_gain);
|
|
|
|
/* compute number of coefficients */
|
|
n = s->coefs_end[bsize] - s->coefs_start;
|
|
for(ch = 0; ch < s->nb_channels; ch++)
|
|
nb_coefs[ch] = n;
|
|
|
|
/* complex coding */
|
|
if (s->use_noise_coding) {
|
|
|
|
for(ch = 0; ch < s->nb_channels; ch++) {
|
|
if (s->channel_coded[ch]) {
|
|
int i, n, a;
|
|
n = s->exponent_high_sizes[bsize];
|
|
for(i=0;i<n;i++) {
|
|
a = get_bits1(&s->gb);
|
|
s->high_band_coded[ch][i] = a;
|
|
/* if noise coding, the coefficients are not transmitted */
|
|
if (a)
|
|
nb_coefs[ch] -= s->exponent_high_bands[bsize][i];
|
|
}
|
|
}
|
|
}
|
|
for(ch = 0; ch < s->nb_channels; ch++) {
|
|
if (s->channel_coded[ch]) {
|
|
int i, n, val, code;
|
|
|
|
n = s->exponent_high_sizes[bsize];
|
|
val = (int)0x80000000;
|
|
for(i=0;i<n;i++) {
|
|
if (s->high_band_coded[ch][i]) {
|
|
if (val == (int)0x80000000) {
|
|
val = get_bits(&s->gb, 7) - 19;
|
|
} else {
|
|
code = get_vlc2(&s->gb, s->hgain_vlc.table, HGAINVLCBITS, HGAINMAX);
|
|
if (code < 0)
|
|
return -1;
|
|
val += code - 18;
|
|
}
|
|
s->high_band_values[ch][i] = val;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
/* exponents can be reused in short blocks. */
|
|
if ((s->block_len_bits == s->frame_len_bits) ||
|
|
get_bits1(&s->gb)) {
|
|
for(ch = 0; ch < s->nb_channels; ch++) {
|
|
if (s->channel_coded[ch]) {
|
|
if (s->use_exp_vlc) {
|
|
if (decode_exp_vlc(s, ch) < 0)
|
|
return -1;
|
|
} else {
|
|
decode_exp_lsp(s, ch);
|
|
}
|
|
s->exponents_bsize[ch] = bsize;
|
|
}
|
|
}
|
|
}
|
|
|
|
/* parse spectral coefficients : just RLE encoding */
|
|
for(ch = 0; ch < s->nb_channels; ch++) {
|
|
if (s->channel_coded[ch]) {
|
|
VLC *coef_vlc;
|
|
int level, run, sign, tindex;
|
|
int16_t *ptr, *eptr;
|
|
const uint16_t *level_table, *run_table;
|
|
|
|
/* special VLC tables are used for ms stereo because
|
|
there is potentially less energy there */
|
|
tindex = (ch == 1 && s->ms_stereo);
|
|
coef_vlc = &s->coef_vlc[tindex];
|
|
run_table = s->run_table[tindex];
|
|
level_table = s->level_table[tindex];
|
|
/* XXX: optimize */
|
|
ptr = &s->coefs1[ch][0];
|
|
eptr = ptr + nb_coefs[ch];
|
|
memset(ptr, 0, s->block_len * sizeof(int16_t));
|
|
for(;;) {
|
|
code = get_vlc2(&s->gb, coef_vlc->table, VLCBITS, VLCMAX);
|
|
if (code < 0)
|
|
return -1;
|
|
if (code == 1) {
|
|
/* EOB */
|
|
break;
|
|
} else if (code == 0) {
|
|
/* escape */
|
|
level = get_bits(&s->gb, coef_nb_bits);
|
|
/* NOTE: this is rather suboptimal. reading
|
|
block_len_bits would be better */
|
|
run = get_bits(&s->gb, s->frame_len_bits);
|
|
} else {
|
|
/* normal code */
|
|
run = run_table[code];
|
|
level = level_table[code];
|
|
}
|
|
sign = get_bits1(&s->gb);
|
|
if (!sign)
|
|
level = -level;
|
|
ptr += run;
|
|
if (ptr >= eptr)
|
|
{
|
|
av_log(NULL, AV_LOG_ERROR, "overflow in spectral RLE, ignoring\n");
|
|
break;
|
|
}
|
|
*ptr++ = level;
|
|
/* NOTE: EOB can be omitted */
|
|
if (ptr >= eptr)
|
|
break;
|
|
}
|
|
}
|
|
if (s->version == 1 && s->nb_channels >= 2) {
|
|
align_get_bits(&s->gb);
|
|
}
|
|
}
|
|
|
|
/* normalize */
|
|
{
|
|
int n4 = s->block_len / 2;
|
|
mdct_norm = 1.0 / (float)n4;
|
|
if (s->version == 1) {
|
|
mdct_norm *= sqrt(n4);
|
|
}
|
|
}
|
|
|
|
/* finally compute the MDCT coefficients */
|
|
for(ch = 0; ch < s->nb_channels; ch++) {
|
|
if (s->channel_coded[ch]) {
|
|
int16_t *coefs1;
|
|
float *coefs, *exponents, mult, mult1, noise;
|
|
int i, j, n, n1, last_high_band, esize;
|
|
float exp_power[HIGH_BAND_MAX_SIZE];
|
|
|
|
coefs1 = s->coefs1[ch];
|
|
exponents = s->exponents[ch];
|
|
esize = s->exponents_bsize[ch];
|
|
mult = pow(10, total_gain * 0.05) / s->max_exponent[ch];
|
|
mult *= mdct_norm;
|
|
coefs = s->coefs[ch];
|
|
if (s->use_noise_coding) {
|
|
mult1 = mult;
|
|
/* very low freqs : noise */
|
|
for(i = 0;i < s->coefs_start; i++) {
|
|
*coefs++ = s->noise_table[s->noise_index] *
|
|
exponents[i<<bsize>>esize] * mult1;
|
|
s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
|
|
}
|
|
|
|
n1 = s->exponent_high_sizes[bsize];
|
|
|
|
/* compute power of high bands */
|
|
exponents = s->exponents[ch] +
|
|
(s->high_band_start[bsize]<<bsize);
|
|
last_high_band = 0; /* avoid warning */
|
|
for(j=0;j<n1;j++) {
|
|
n = s->exponent_high_bands[s->frame_len_bits -
|
|
s->block_len_bits][j];
|
|
if (s->high_band_coded[ch][j]) {
|
|
float e2, v;
|
|
e2 = 0;
|
|
for(i = 0;i < n; i++) {
|
|
v = exponents[i<<bsize>>esize];
|
|
e2 += v * v;
|
|
}
|
|
exp_power[j] = e2 / n;
|
|
last_high_band = j;
|
|
tprintf(s->avctx, "%d: power=%f (%d)\n", j, exp_power[j], n);
|
|
}
|
|
exponents += n<<bsize;
|
|
}
|
|
|
|
/* main freqs and high freqs */
|
|
exponents = s->exponents[ch] + (s->coefs_start<<bsize);
|
|
for(j=-1;j<n1;j++) {
|
|
if (j < 0) {
|
|
n = s->high_band_start[bsize] -
|
|
s->coefs_start;
|
|
} else {
|
|
n = s->exponent_high_bands[s->frame_len_bits -
|
|
s->block_len_bits][j];
|
|
}
|
|
if (j >= 0 && s->high_band_coded[ch][j]) {
|
|
/* use noise with specified power */
|
|
mult1 = sqrt(exp_power[j] / exp_power[last_high_band]);
|
|
/* XXX: use a table */
|
|
mult1 = mult1 * pow(10, s->high_band_values[ch][j] * 0.05);
|
|
mult1 = mult1 / (s->max_exponent[ch] * s->noise_mult);
|
|
mult1 *= mdct_norm;
|
|
for(i = 0;i < n; i++) {
|
|
noise = s->noise_table[s->noise_index];
|
|
s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
|
|
*coefs++ = noise *
|
|
exponents[i<<bsize>>esize] * mult1;
|
|
}
|
|
exponents += n<<bsize;
|
|
} else {
|
|
/* coded values + small noise */
|
|
for(i = 0;i < n; i++) {
|
|
noise = s->noise_table[s->noise_index];
|
|
s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
|
|
*coefs++ = ((*coefs1++) + noise) *
|
|
exponents[i<<bsize>>esize] * mult;
|
|
}
|
|
exponents += n<<bsize;
|
|
}
|
|
}
|
|
|
|
/* very high freqs : noise */
|
|
n = s->block_len - s->coefs_end[bsize];
|
|
mult1 = mult * exponents[((-1<<bsize))>>esize];
|
|
for(i = 0; i < n; i++) {
|
|
*coefs++ = s->noise_table[s->noise_index] * mult1;
|
|
s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
|
|
}
|
|
} else {
|
|
/* XXX: optimize more */
|
|
for(i = 0;i < s->coefs_start; i++)
|
|
*coefs++ = 0.0;
|
|
n = nb_coefs[ch];
|
|
for(i = 0;i < n; i++) {
|
|
*coefs++ = coefs1[i] * exponents[i<<bsize>>esize] * mult;
|
|
}
|
|
n = s->block_len - s->coefs_end[bsize];
|
|
for(i = 0;i < n; i++)
|
|
*coefs++ = 0.0;
|
|
}
|
|
}
|
|
}
|
|
|
|
#ifdef TRACE
|
|
for(ch = 0; ch < s->nb_channels; ch++) {
|
|
if (s->channel_coded[ch]) {
|
|
dump_floats(s, "exponents", 3, s->exponents[ch], s->block_len);
|
|
dump_floats(s, "coefs", 1, s->coefs[ch], s->block_len);
|
|
}
|
|
}
|
|
#endif
|
|
|
|
if (s->ms_stereo && s->channel_coded[1]) {
|
|
float a, b;
|
|
int i;
|
|
|
|
/* nominal case for ms stereo: we do it before mdct */
|
|
/* no need to optimize this case because it should almost
|
|
never happen */
|
|
if (!s->channel_coded[0]) {
|
|
tprintf(s->avctx, "rare ms-stereo case happened\n");
|
|
memset(s->coefs[0], 0, sizeof(float) * s->block_len);
|
|
s->channel_coded[0] = 1;
|
|
}
|
|
|
|
for(i = 0; i < s->block_len; i++) {
|
|
a = s->coefs[0][i];
|
|
b = s->coefs[1][i];
|
|
s->coefs[0][i] = a + b;
|
|
s->coefs[1][i] = a - b;
|
|
}
|
|
}
|
|
|
|
for(ch = 0; ch < s->nb_channels; ch++) {
|
|
if (s->channel_coded[ch]) {
|
|
int n4, index, n;
|
|
|
|
n = s->block_len;
|
|
n4 = s->block_len / 2;
|
|
s->mdct_ctx[bsize].fft.imdct_calc(&s->mdct_ctx[bsize],
|
|
s->output, s->coefs[ch], s->mdct_tmp);
|
|
|
|
/* multiply by the window and add in the frame */
|
|
index = (s->frame_len / 2) + s->block_pos - n4;
|
|
wma_window(s, &s->frame_out[ch][index]);
|
|
|
|
/* specific fast case for ms-stereo : add to second
|
|
channel if it is not coded */
|
|
if (s->ms_stereo && !s->channel_coded[1]) {
|
|
wma_window(s, &s->frame_out[1][index]);
|
|
}
|
|
}
|
|
}
|
|
next:
|
|
/* update block number */
|
|
s->block_num++;
|
|
s->block_pos += s->block_len;
|
|
if (s->block_pos >= s->frame_len)
|
|
return 1;
|
|
else
|
|
return 0;
|
|
}
|
|
|
|
/* decode a frame of frame_len samples */
|
|
static int wma_decode_frame(WMACodecContext *s, int16_t *samples)
|
|
{
|
|
int ret, i, n, ch, incr;
|
|
int16_t *ptr;
|
|
float *iptr;
|
|
|
|
#ifdef TRACE
|
|
tprintf(s->avctx, "***decode_frame: %d size=%d\n", s->frame_count++, s->frame_len);
|
|
#endif
|
|
|
|
/* read each block */
|
|
s->block_num = 0;
|
|
s->block_pos = 0;
|
|
for(;;) {
|
|
ret = wma_decode_block(s);
|
|
if (ret < 0)
|
|
return -1;
|
|
if (ret)
|
|
break;
|
|
}
|
|
|
|
/* convert frame to integer */
|
|
n = s->frame_len;
|
|
incr = s->nb_channels;
|
|
for(ch = 0; ch < s->nb_channels; ch++) {
|
|
ptr = samples + ch;
|
|
iptr = s->frame_out[ch];
|
|
|
|
for(i=0;i<n;i++) {
|
|
*ptr = av_clip_int16(lrintf(*iptr++));
|
|
ptr += incr;
|
|
}
|
|
/* prepare for next block */
|
|
memmove(&s->frame_out[ch][0], &s->frame_out[ch][s->frame_len],
|
|
s->frame_len * sizeof(float));
|
|
}
|
|
|
|
#ifdef TRACE
|
|
dump_shorts(s, "samples", samples, n * s->nb_channels);
|
|
#endif
|
|
return 0;
|
|
}
|
|
|
|
static int wma_decode_superframe(AVCodecContext *avctx,
|
|
void *data, int *data_size,
|
|
const uint8_t *buf, int buf_size)
|
|
{
|
|
WMACodecContext *s = avctx->priv_data;
|
|
int nb_frames, bit_offset, i, pos, len;
|
|
uint8_t *q;
|
|
int16_t *samples;
|
|
|
|
tprintf(avctx, "***decode_superframe:\n");
|
|
|
|
if(buf_size==0){
|
|
s->last_superframe_len = 0;
|
|
return 0;
|
|
}
|
|
if (buf_size < s->block_align)
|
|
return 0;
|
|
buf_size = s->block_align;
|
|
|
|
samples = data;
|
|
|
|
init_get_bits(&s->gb, buf, buf_size*8);
|
|
|
|
if (s->use_bit_reservoir) {
|
|
/* read super frame header */
|
|
skip_bits(&s->gb, 4); /* super frame index */
|
|
nb_frames = get_bits(&s->gb, 4) - 1;
|
|
|
|
bit_offset = get_bits(&s->gb, s->byte_offset_bits + 3);
|
|
|
|
if (s->last_superframe_len > 0) {
|
|
// printf("skip=%d\n", s->last_bitoffset);
|
|
/* add bit_offset bits to last frame */
|
|
if ((s->last_superframe_len + ((bit_offset + 7) >> 3)) >
|
|
MAX_CODED_SUPERFRAME_SIZE)
|
|
goto fail;
|
|
q = s->last_superframe + s->last_superframe_len;
|
|
len = bit_offset;
|
|
while (len > 7) {
|
|
*q++ = (get_bits)(&s->gb, 8);
|
|
len -= 8;
|
|
}
|
|
if (len > 0) {
|
|
*q++ = (get_bits)(&s->gb, len) << (8 - len);
|
|
}
|
|
|
|
/* XXX: bit_offset bits into last frame */
|
|
init_get_bits(&s->gb, s->last_superframe, MAX_CODED_SUPERFRAME_SIZE*8);
|
|
/* skip unused bits */
|
|
if (s->last_bitoffset > 0)
|
|
skip_bits(&s->gb, s->last_bitoffset);
|
|
/* this frame is stored in the last superframe and in the
|
|
current one */
|
|
if (wma_decode_frame(s, samples) < 0)
|
|
goto fail;
|
|
samples += s->nb_channels * s->frame_len;
|
|
}
|
|
|
|
/* read each frame starting from bit_offset */
|
|
pos = bit_offset + 4 + 4 + s->byte_offset_bits + 3;
|
|
init_get_bits(&s->gb, buf + (pos >> 3), (MAX_CODED_SUPERFRAME_SIZE - (pos >> 3))*8);
|
|
len = pos & 7;
|
|
if (len > 0)
|
|
skip_bits(&s->gb, len);
|
|
|
|
s->reset_block_lengths = 1;
|
|
for(i=0;i<nb_frames;i++) {
|
|
if (wma_decode_frame(s, samples) < 0)
|
|
goto fail;
|
|
samples += s->nb_channels * s->frame_len;
|
|
}
|
|
|
|
/* we copy the end of the frame in the last frame buffer */
|
|
pos = get_bits_count(&s->gb) + ((bit_offset + 4 + 4 + s->byte_offset_bits + 3) & ~7);
|
|
s->last_bitoffset = pos & 7;
|
|
pos >>= 3;
|
|
len = buf_size - pos;
|
|
if (len > MAX_CODED_SUPERFRAME_SIZE || len < 0) {
|
|
goto fail;
|
|
}
|
|
s->last_superframe_len = len;
|
|
memcpy(s->last_superframe, buf + pos, len);
|
|
} else {
|
|
/* single frame decode */
|
|
if (wma_decode_frame(s, samples) < 0)
|
|
goto fail;
|
|
samples += s->nb_channels * s->frame_len;
|
|
}
|
|
|
|
//av_log(NULL, AV_LOG_ERROR, "%d %d %d %d outbytes:%d eaten:%d\n", s->frame_len_bits, s->block_len_bits, s->frame_len, s->block_len, (int8_t *)samples - (int8_t *)data, s->block_align);
|
|
|
|
*data_size = (int8_t *)samples - (int8_t *)data;
|
|
return s->block_align;
|
|
fail:
|
|
/* when error, we reset the bit reservoir */
|
|
s->last_superframe_len = 0;
|
|
return -1;
|
|
}
|
|
|
|
AVCodec wmav1_decoder =
|
|
{
|
|
"wmav1",
|
|
CODEC_TYPE_AUDIO,
|
|
CODEC_ID_WMAV1,
|
|
sizeof(WMACodecContext),
|
|
wma_decode_init,
|
|
NULL,
|
|
ff_wma_end,
|
|
wma_decode_superframe,
|
|
.long_name = "Windows Media Audio 1",
|
|
};
|
|
|
|
AVCodec wmav2_decoder =
|
|
{
|
|
"wmav2",
|
|
CODEC_TYPE_AUDIO,
|
|
CODEC_ID_WMAV2,
|
|
sizeof(WMACodecContext),
|
|
wma_decode_init,
|
|
NULL,
|
|
ff_wma_end,
|
|
wma_decode_superframe,
|
|
.long_name = "Windows Media Audio 2",
|
|
};
|