virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2025-01-24 13:56:33 +02:00
Code Issues Releases Activity

avfilter/af_adynamicequalizer: add precision option

Browse Source
This commit is contained in:
Paul B Mahol 2023-04-29 10:40:18 +02:00
parent 41dd50ad0d
commit f247a3d82d
3 changed files with 347 additions and 217 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
doc/filters.texi
View File

@ -1107,6 +1107,20 @@ Stop picking threshold value.
@item on
Start picking threshold value.
@end table
@item precision
Set which precision to use when processing samples.
@table @option
@item auto
Auto pick internal sample format depending on other filters.
@item float
Always use single-floating point precision sample format.
@item double
Always use double-floating point precision sample format.
@end table
@end table
@subsection Commands

270
libavfilter/adynamicequalizer_template.c Normal file
View File

@ -0,0 +1,270 @@
/*
* 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
*/
#undef ftype
#undef SQRT
#undef TAN
#undef ONE
#undef TWO
#undef ZERO
#undef FMAX
#undef FMIN
#undef CLIP
#undef SAMPLE_FORMAT
#undef FABS
#if DEPTH == 32
#define SAMPLE_FORMAT float
#define SQRT sqrtf
#define TAN tanf
#define ONE 1.f
#define TWO 2.f
#define ZERO 0.f
#define FMIN fminf
#define FMAX fmaxf
#define CLIP av_clipf
#define FABS fabsf
#define ftype float
#else
#define SAMPLE_FORMAT double
#define SQRT sqrt
#define TAN tan
#define ONE 1.0
#define TWO 2.0
#define ZERO 0.0
#define FMIN fmin
#define FMAX fmax
#define CLIP av_clipd
#define FABS fabs
#define ftype double
#endif
#define fn3(a,b) a##_##b
#define fn2(a,b) fn3(a,b)
#define fn(a) fn2(a, SAMPLE_FORMAT)
static ftype fn(get_svf)(ftype in, const ftype *m, const ftype *a, ftype *b)
{
const ftype v0 = in;
const ftype v3 = v0 - b[1];
const ftype v1 = a[0] * b[0] + a[1] * v3;
const ftype v2 = b[1] + a[1] * b[0] + a[2] * v3;
b[0] = TWO * v1 - b[0];
b[1] = TWO * v2 - b[1];
return m[0] * v0 + m[1] * v1 + m[2] * v2;
}
static int fn(filter_prepare)(AVFilterContext *ctx)
{
AudioDynamicEqualizerContext *s = ctx->priv;
const ftype sample_rate = ctx->inputs[0]->sample_rate;
const ftype dfrequency = FMIN(s->dfrequency, sample_rate * 0.5);
const ftype dg = TAN(M_PI * dfrequency / sample_rate);
const ftype dqfactor = s->dqfactor;
const int dftype = s->dftype;
ftype *da = fn(s->da);
ftype *dm = fn(s->dm);
ftype k;
s->attack_coef = get_coef(s->attack, sample_rate);
s->release_coef = get_coef(s->release, sample_rate);
switch (dftype) {
case 0:
k = ONE / dqfactor;
da[0] = ONE / (ONE + dg * (dg + k));
da[1] = dg * da[0];
da[2] = dg * da[1];
dm[0] = ZERO;
dm[1] = k;
dm[2] = ZERO;
break;
case 1:
k = ONE / dqfactor;
da[0] = ONE / (ONE + dg * (dg + k));
da[1] = dg * da[0];
da[2] = dg * da[1];
dm[0] = ZERO;
dm[1] = ZERO;
dm[2] = ONE;
break;
case 2:
k = ONE / dqfactor;
da[0] = ONE / (ONE + dg * (dg + k));
da[1] = dg * da[0];
da[2] = dg * da[1];
dm[0] = ZERO;
dm[1] = -k;
dm[2] = -ONE;
break;
case 3:
k = ONE / dqfactor;
da[0] = ONE / (ONE + dg * (dg + k));
da[1] = dg * da[0];
da[2] = dg * da[1];
dm[0] = ZERO;
dm[1] = -k;
dm[2] = -TWO;
break;
}
return 0;
}
static int fn(filter_channels)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
{
AudioDynamicEqualizerContext *s = ctx->priv;
ThreadData *td = arg;
AVFrame *in = td->in;
AVFrame *out = td->out;
const ftype sample_rate = in->sample_rate;
const ftype makeup = s->makeup;
const ftype ratio = s->ratio;
const ftype range = s->range;
const ftype tfrequency = FMIN(s->tfrequency, sample_rate * 0.5);
const ftype release = s->release_coef;
const ftype irelease = ONE - release;
const ftype attack = s->attack_coef;
const ftype iattack = ONE - attack;
const ftype tqfactor = s->tqfactor;
const ftype itqfactor = ONE / tqfactor;
const ftype fg = TAN(M_PI * tfrequency / sample_rate);
const int start = (in->ch_layout.nb_channels * jobnr) / nb_jobs;
const int end = (in->ch_layout.nb_channels * (jobnr+1)) / nb_jobs;
const int detection = s->detection;
const int direction = s->direction;
const int tftype = s->tftype;
const int mode = s->mode;
const ftype *da = fn(s->da);
const ftype *dm = fn(s->dm);
for (int ch = start; ch < end; ch++) {
const ftype *src = (const ftype *)in->extended_data[ch];
ftype *dst = (ftype *)out->extended_data[ch];
ftype *state = (ftype *)s->state->extended_data[ch];
const ftype threshold = detection == 0 ? state[5] : s->threshold;
if (detection < 0)
state[5] = threshold;
for (int n = 0; n < out->nb_samples; n++) {
ftype detect, gain, v, listen;
ftype fa[3], fm[3];
ftype k, g;
detect = listen = fn(get_svf)(src[n], dm, da, state);
detect = FABS(detect);
if (detection > 0)
state[5] = FMAX(state[5], detect);
if (direction == 0) {
if (detect < threshold) {
if (mode == 0)
detect = ONE / CLIP(ONE + makeup + (threshold - detect) * ratio, ONE, range);
else
detect = CLIP(ONE + makeup + (threshold - detect) * ratio, ONE, range);
} else {
detect = ONE;
}
} else {
if (detect > threshold) {
if (mode == 0)
detect = ONE / CLIP(ONE + makeup + (detect - threshold) * ratio, ONE, range);
else
detect = CLIP(ONE + makeup + (detect - threshold) * ratio, ONE, range);
} else {
detect = ONE;
}
}
if (direction == 0) {
if (detect > state[4]) {
detect = iattack * detect + attack * state[4];
} else {
detect = irelease * detect + release * state[4];
}
} else {
if (detect < state[4]) {
detect = iattack * detect + attack * state[4];
} else {
detect = irelease * detect + release * state[4];
}
}
if (state[4] != detect || n == 0) {
state[4] = gain = detect;
switch (tftype) {
case 0:
k = ONE / (tqfactor * gain);
fa[0] = ONE / (ONE + fg * (fg + k));
fa[1] = fg * fa[0];
fa[2] = fg * fa[1];
fm[0] = ONE;
fm[1] = k * (gain * gain - ONE);
fm[2] = ZERO;
break;
case 1:
k = itqfactor;
g = fg / SQRT(gain);
fa[0] = ONE / (ONE + g * (g + k));
fa[1] = g * fa[0];
fa[2] = g * fa[1];
fm[0] = ONE;
fm[1] = k * (gain - ONE);
fm[2] = gain * gain - ONE;
break;
case 2:
k = itqfactor;
g = fg / SQRT(gain);
fa[0] = ONE / (ONE + g * (g + k));
fa[1] = g * fa[0];
fa[2] = g * fa[1];
fm[0] = gain * gain;
fm[1] = k * (ONE - gain) * gain;
fm[2] = ONE - gain * gain;
break;
}
}
v = fn(get_svf)(src[n], fm, fa, &state[2]);
v = mode == -1 ? listen : v;
dst[n] = ctx->is_disabled ? src[n] : v;
}
}
return 0;
}

280
libavfilter/af_adynamicequalizer.c
View File

@ -43,242 +43,82 @@ typedef struct AudioDynamicEqualizerContext {
int detection;
int tftype;
int dftype;
int precision;
int format;
int (*filter_prepare)(AVFilterContext *ctx);
int (*filter_channels)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
double da_double[3], dm_double[3];
float da_float[3], dm_float[3];
double da[3], dm[3];
AVFrame *state;
} AudioDynamicEqualizerContext;
static int config_input(AVFilterLink *inlink)
static int query_formats(AVFilterContext *ctx)
{
AVFilterContext *ctx = inlink->dst;
AudioDynamicEqualizerContext *s = ctx->priv;
static const enum AVSampleFormat sample_fmts[3][3] = {
{ AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_DBLP, AV_SAMPLE_FMT_NONE },
{ AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE },
{ AV_SAMPLE_FMT_DBLP, AV_SAMPLE_FMT_NONE },
};
int ret;
s->state = ff_get_audio_buffer(inlink, 8);
if (!s->state)
return AVERROR(ENOMEM);
if ((ret = ff_set_common_all_channel_counts(ctx)) < 0)
return ret;
for (int ch = 0; ch < inlink->ch_layout.nb_channels; ch++) {
double *state = (double *)s->state->extended_data[ch];
if ((ret = ff_set_common_formats_from_list(ctx, sample_fmts[s->precision])) < 0)
return ret;
state[4] = 1.;
}
return 0;
return ff_set_common_all_samplerates(ctx);
}
static double get_svf(double in, const double *m, const double *a, double *b)
{
const double v0 = in;
const double v3 = v0 - b[1];
const double v1 = a[0] * b[0] + a[1] * v3;
const double v2 = b[1] + a[1] * b[0] + a[2] * v3;
b[0] = 2. * v1 - b[0];
b[1] = 2. * v2 - b[1];
return m[0] * v0 + m[1] * v1 + m[2] * v2;
}
typedef struct ThreadData {
AVFrame *in, *out;
} ThreadData;
static double get_coef(double x, double sr)
{
return exp(-1000. / (x * sr));
}
static int filter_prepare(AVFilterContext *ctx)
typedef struct ThreadData {
AVFrame *in, *out;
} ThreadData;
#define DEPTH 32
#include "adynamicequalizer_template.c"
#undef DEPTH
#define DEPTH 64
#include "adynamicequalizer_template.c"
static int config_input(AVFilterLink *inlink)
{
AVFilterContext *ctx = inlink->dst;
AudioDynamicEqualizerContext *s = ctx->priv;
const double sample_rate = ctx->inputs[0]->sample_rate;
const double dfrequency = fmin(s->dfrequency, sample_rate * 0.5);
const double dg = tan(M_PI * dfrequency / sample_rate);
const double dqfactor = s->dqfactor;
const int dftype = s->dftype;
double *da = s->da;
double *dm = s->dm;
double k;
s->attack_coef = get_coef(s->attack, sample_rate);
s->release_coef = get_coef(s->release, sample_rate);
s->format = inlink->format;
s->state = ff_get_audio_buffer(inlink, 8);
if (!s->state)
return AVERROR(ENOMEM);
switch (dftype) {
case 0:
k = 1. / dqfactor;
switch (s->format) {
case AV_SAMPLE_FMT_DBLP:
for (int ch = 0; ch < inlink->ch_layout.nb_channels; ch++) {
double *state = (double *)s->state->extended_data[ch];
da[0] = 1. / (1. + dg * (dg + k));
da[1] = dg * da[0];
da[2] = dg * da[1];
dm[0] = 0.;
dm[1] = k;
dm[2] = 0.;
break;
case 1:
k = 1. / dqfactor;
da[0] = 1. / (1. + dg * (dg + k));
da[1] = dg * da[0];
da[2] = dg * da[1];
dm[0] = 0.;
dm[1] = 0.;
dm[2] = 1.;
break;
case 2:
k = 1. / dqfactor;
da[0] = 1. / (1. + dg * (dg + k));
da[1] = dg * da[0];
da[2] = dg * da[1];
dm[0] = 0.;
dm[1] = -k;
dm[2] = -1.;
break;
case 3:
k = 1. / dqfactor;
da[0] = 1. / (1. + dg * (dg + k));
da[1] = dg * da[0];
da[2] = dg * da[1];
dm[0] = 0.;
dm[1] = -k;
dm[2] = -2.;
break;
}
return 0;
}
static int filter_channels(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
{
AudioDynamicEqualizerContext *s = ctx->priv;
ThreadData *td = arg;
AVFrame *in = td->in;
AVFrame *out = td->out;
const double sample_rate = in->sample_rate;
const double makeup = s->makeup;
const double ratio = s->ratio;
const double range = s->range;
const double tfrequency = fmin(s->tfrequency, sample_rate * 0.5);
const double release = s->release_coef;
const double irelease = 1. - release;
const double attack = s->attack_coef;
const double iattack = 1. - attack;
const double tqfactor = s->tqfactor;
const double fg = tan(M_PI * tfrequency / sample_rate);
const int start = (in->ch_layout.nb_channels * jobnr) / nb_jobs;
const int end = (in->ch_layout.nb_channels * (jobnr+1)) / nb_jobs;
const int detection = s->detection;
const int direction = s->direction;
const int tftype = s->tftype;
const int mode = s->mode;
const double *da = s->da;
const double *dm = s->dm;
for (int ch = start; ch < end; ch++) {
const double *src = (const double *)in->extended_data[ch];
double *dst = (double *)out->extended_data[ch];
double *state = (double *)s->state->extended_data[ch];
const double threshold = detection == 0 ? state[5] : s->threshold;
if (detection < 0)
state[5] = threshold;
for (int n = 0; n < out->nb_samples; n++) {
double detect, gain, v, listen;
double fa[3], fm[3];
double k, g;
detect = listen = get_svf(src[n], dm, da, state);
detect = fabs(detect);
if (detection > 0)
state[5] = fmax(state[5], detect);
if (direction == 0) {
if (detect < threshold) {
if (mode == 0)
detect = 1. / av_clipd(1. + makeup + (threshold - detect) * ratio, 1., range);
else
detect = av_clipd(1. + makeup + (threshold - detect) * ratio, 1., range);
} else {
detect = 1.;
}
} else {
if (detect > threshold) {
if (mode == 0)
detect = 1. / av_clipd(1. + makeup + (detect - threshold) * ratio, 1., range);
else
detect = av_clipd(1. + makeup + (detect - threshold) * ratio, 1., range);
} else {
detect = 1.;
}
}
if (direction == 0) {
if (detect > state[4]) {
detect = iattack * detect + attack * state[4];
} else {
detect = irelease * detect + release * state[4];
}
} else {
if (detect < state[4]) {
detect = iattack * detect + attack * state[4];
} else {
detect = irelease * detect + release * state[4];
}
}
if (state[4] != detect || n == 0) {
state[4] = gain = detect;
switch (tftype) {
case 0:
k = 1. / (tqfactor * gain);
fa[0] = 1. / (1. + fg * (fg + k));
fa[1] = fg * fa[0];
fa[2] = fg * fa[1];
fm[0] = 1.;
fm[1] = k * (gain * gain - 1.);
fm[2] = 0.;
break;
case 1:
k = 1. / tqfactor;
g = fg / sqrt(gain);
fa[0] = 1. / (1. + g * (g + k));
fa[1] = g * fa[0];
fa[2] = g * fa[1];
fm[0] = 1.;
fm[1] = k * (gain - 1.);
fm[2] = gain * gain - 1.;
break;
case 2:
k = 1. / tqfactor;
g = fg / sqrt(gain);
fa[0] = 1. / (1. + g * (g + k));
fa[1] = g * fa[0];
fa[2] = g * fa[1];
fm[0] = gain * gain;
fm[1] = k * (1. - gain) * gain;
fm[2] = 1. - gain * gain;
break;
}
}
v = get_svf(src[n], fm, fa, &state[2]);
v = mode == -1 ? listen : v;
dst[n] = ctx->is_disabled ? src[n] : v;
state[4] = 1.;
}
s->filter_prepare = filter_prepare_double;
s->filter_channels = filter_channels_double;
break;
case AV_SAMPLE_FMT_FLTP:
for (int ch = 0; ch < inlink->ch_layout.nb_channels; ch++) {
float *state = (float *)s->state->extended_data[ch];
state[4] = 1.;
}
s->filter_prepare = filter_prepare_float;
s->filter_channels = filter_channels_float;
break;
}
return 0;
@ -288,6 +128,7 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
{
AVFilterContext *ctx = inlink->dst;
AVFilterLink *outlink = ctx->outputs[0];
AudioDynamicEqualizerContext *s = ctx->priv;
ThreadData td;
AVFrame *out;
@ -304,8 +145,8 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
td.in = in;
td.out = out;
filter_prepare(ctx);
ff_filter_execute(ctx, filter_channels, &td, NULL,
s->filter_prepare(ctx);
ff_filter_execute(ctx, s->filter_channels, &td, NULL,
FFMIN(outlink->ch_layout.nb_channels, ff_filter_get_nb_threads(ctx)));
if (out != in)
@ -321,6 +162,7 @@ static av_cold void uninit(AVFilterContext *ctx)
}
#define OFFSET(x) offsetof(AudioDynamicEqualizerContext, x)
#define AF AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
#define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
static const AVOption adynamicequalizer_options[] = {
@ -354,6 +196,10 @@ static const AVOption adynamicequalizer_options[] = {
{ "disabled", 0, 0, AV_OPT_TYPE_CONST, {.i64=-1}, 0, 0, FLAGS, "auto" },
{ "off", 0, 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, FLAGS, "auto" },
{ "on", 0, 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, FLAGS, "auto" },
{ "precision", "set processing precision", OFFSET(precision), AV_OPT_TYPE_INT, {.i64=0}, 0, 2, AF, "precision" },
{ "auto", "set auto processing precision", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, "precision" },
{ "float", "set single-floating point processing precision", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF, "precision" },
{ "double","set double-floating point processing precision", 0, AV_OPT_TYPE_CONST, {.i64=2}, 0, 0, AF, "precision" },
{ NULL }
};
@ -383,7 +229,7 @@ const AVFilter ff_af_adynamicequalizer = {
.uninit = uninit,
FILTER_INPUTS(inputs),
FILTER_OUTPUTS(outputs),
FILTER_SINGLE_SAMPLEFMT(AV_SAMPLE_FMT_DBLP),
FILTER_QUERY_FUNC(query_formats),
.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL |
AVFILTER_FLAG_SLICE_THREADS,
.process_command = ff_filter_process_command,