[FFmpeg-devel] [PATCH] avfilter: add audio dynamic equalizer filter
Paul B Mahol
onemda at gmail.com
Sun Nov 28 15:21:08 EET 2021
Signed-off-by: Paul B Mahol <onemda at gmail.com>
---
doc/filters.texi | 75 +++++++
libavfilter/Makefile | 1 +
libavfilter/af_adynamicequalizer.c | 310 +++++++++++++++++++++++++++++
libavfilter/allfilters.c | 1 +
4 files changed, 387 insertions(+)
create mode 100644 libavfilter/af_adynamicequalizer.c
diff --git a/doc/filters.texi b/doc/filters.texi
index 7852948d2f..cfddefff30 100644
--- a/doc/filters.texi
+++ b/doc/filters.texi
@@ -843,6 +843,81 @@ Compute derivative/integral of audio stream.
Applying both filters one after another produces original audio.
+ at section adynamicequalizer
+
+Apply dynamic equalization to input audio stream.
+
+A description of the accepted options follows.
+
+ at table @option
+ at item threshold
+Set the detection threshold used to trigger equalization.
+Threshold detection is using bandpass filter.
+Default value is 0. Allowed range is from 0 to 50.
+
+ at item dfrequency
+Set the detection frequency in Hz used for bandpass filter used to trigger equalization.
+Default value is 1000 Hz. Allowed range is between 2 and 1000000 Hz.
+
+ at item dqfactor
+Set the detection resonance factor for bandpass filter used to trigger equalization.
+Default value is 1. Allowed range is from 0.001 to 1000.
+
+ at item sfrequency
+Set the frequency of lowpass filter to smooth equalization filter gains.
+Default value is 100 Hz. Allowed range is between 2 and 1000000 Hz.
+
+ at item sqfactor
+Set the resonance factor of lowpass filter to smooth equalization filter gains.
+Default value is 1. Allowed range is from 0.001 to 1000.
+
+ at item tfrequency
+Set the target frequency of equalization filter.
+Default value is 1000 Hz. Allowed range is between 2 and 1000000 Hz.
+
+ at item tqfactor
+Set the target resonance factor for target equalization filter.
+Default value is 1. Allowed range is from 0.001 to 1000.
+
+ at item attack
+Set the amount of milliseconds the signal from detection has to rise above
+the detection threshold before equalization starts.
+Default is 10. Allowed range is between 1 and 2000.
+
+ at item release
+Set the amount of milliseconds the signal from detection has to fall below the
+detection threshold before equalization ends.
+Default is 80. Allowed range is between 1 and 2000.
+
+ at item knee
+Curve the sharp knee around the detection threshold to calculate
+equalization gain more softly.
+Default is 2. Allowed range is between 1 and 8.
+
+ at item ratio
+Set the ratio by which the equalization gain is raised.
+Default is 1. Allowed range is between 1 and 20.
+
+ at item range
+Set the max allowed cut/boost amount. Default is 0.06125.
+Allowed range is from 0.00000001 to 1.
+
+ at item mode
+Set the mode of filter operation, can be one of the following:
+
+ at table @samp
+ at item cut
+Cut frequencies above detection threshold.
+ at item boost
+Boost frequencies bellow detection threshold.
+ at end table
+Default mode is @samp{cut}.
+ at end table
+
+ at subsection Commands
+
+This filter supports the all above options as @ref{commands}.
+
@section adynamicsmooth
Apply dynamic smoothing to input audio stream.
diff --git a/libavfilter/Makefile b/libavfilter/Makefile
index c8082c4a2f..d40be4b252 100644
--- a/libavfilter/Makefile
+++ b/libavfilter/Makefile
@@ -44,6 +44,7 @@ OBJS-$(CONFIG_ADECORRELATE_FILTER) += af_adecorrelate.o
OBJS-$(CONFIG_ADELAY_FILTER) += af_adelay.o
OBJS-$(CONFIG_ADENORM_FILTER) += af_adenorm.o
OBJS-$(CONFIG_ADERIVATIVE_FILTER) += af_aderivative.o
+OBJS-$(CONFIG_ADYNAMICEQUALIZER_FILTER) += af_adynamicequalizer.o
OBJS-$(CONFIG_ADYNAMICSMOOTH_FILTER) += af_adynamicsmooth.o
OBJS-$(CONFIG_AECHO_FILTER) += af_aecho.o
OBJS-$(CONFIG_AEMPHASIS_FILTER) += af_aemphasis.o
diff --git a/libavfilter/af_adynamicequalizer.c b/libavfilter/af_adynamicequalizer.c
new file mode 100644
index 0000000000..da3d96d7cf
--- /dev/null
+++ b/libavfilter/af_adynamicequalizer.c
@@ -0,0 +1,310 @@
+/*
+ * 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
+ */
+
+#include "libavutil/ffmath.h"
+#include "libavutil/opt.h"
+#include "avfilter.h"
+#include "audio.h"
+#include "formats.h"
+#include "hermite.h"
+
+typedef struct AudioDynamicEqualizerContext {
+ const AVClass *class;
+
+ double threshold;
+ double lin_threshold;
+ double dfrequency;
+ double dqfactor;
+ double tfrequency;
+ double tqfactor;
+ double sfrequency;
+ double sqfactor;
+ double attack;
+ double release;
+ double knee;
+ double ratio;
+ double range;
+ double knee_sqrt;
+ double attack_coeff;
+ double release_coeff;
+ double lin_knee_start;
+ double lin_knee_stop;
+ double knee_start;
+ double knee_stop;
+ double compressed_knee_start;
+ int mode;
+
+ AVFrame *state;
+} AudioDynamicEqualizerContext;
+
+static int config_input(AVFilterLink *inlink)
+{
+ AVFilterContext *ctx = inlink->dst;
+ AudioDynamicEqualizerContext *s = ctx->priv;
+
+ s->state = ff_get_audio_buffer(inlink, 7);
+ if (!s->state)
+ return AVERROR(ENOMEM);
+
+ return 0;
+}
+
+static double get_gain(AVFilterContext *ctx, double in,
+ double sample_rate, int ch)
+{
+ AudioDynamicEqualizerContext *s = ctx->priv;
+ double *state = (double *)s->state->extended_data[ch];
+ const double ratio = s->ratio;
+ const double knee = s->knee;
+ const double attack_coeff = s->attack_coeff;
+ const double release_coeff = s->release_coeff;
+ const double lin_knee_start = s->lin_knee_start;
+ const double threshold = s->threshold;
+ const double knee_start = s->knee_start;
+ const double knee_stop = s->knee_stop;
+ const double compressed_knee_start = s->compressed_knee_start;
+ const double abs_sample = fabs(in);
+ const int mode = s->mode;
+ double lin_slope = state[2];
+ double range = s->range;
+ double gain = 0.;
+ double slope;
+ int detected;
+
+ lin_slope += (abs_sample - lin_slope) * (abs_sample > lin_slope ? attack_coeff : release_coeff);
+
+ detected = lin_slope > lin_knee_start;
+
+ state[2] = lin_slope;
+ if (lin_slope <= 0. || !detected)
+ return 1.;
+
+ slope = log(lin_slope);
+ gain = (slope - threshold) * ratio + threshold;
+
+ if (knee >= 1.)
+ gain = hermite_interpolation(slope, knee_stop, knee_start,
+ knee_stop, compressed_knee_start,
+ 1., ratio);
+ if (!mode)
+ gain = -gain;
+
+ gain = exp(gain - slope);
+ gain = mode ? av_clipd(gain, 1., 1. / range) : av_clipd(gain, range, 1.);
+
+ return gain;
+}
+
+static double get_svf(double in, double *m, 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 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 range = s->range;
+ const double irange = 1. / range;
+ const double dfrequency = fmin(s->dfrequency, sample_rate * 0.5);
+ const double tfrequency = fmin(s->tfrequency, sample_rate * 0.5);
+ const double sfrequency = fmin(s->sfrequency, sample_rate * 0.5);
+ const double dqfactor = s->dqfactor;
+ const double tqfactor = s->tqfactor;
+ const double sqfactor = s->tqfactor;
+ const double fg = tan(M_PI * tfrequency / sample_rate);
+ const double dg = tan(M_PI * dfrequency / sample_rate);
+ const double sg = tan(M_PI * sfrequency / sample_rate);
+ const int start = (in->channels * jobnr) / nb_jobs;
+ const int end = (in->channels * (jobnr+1)) / nb_jobs;
+ const int mode = s->mode;
+ double da[3], dm[3];
+
+ {
+ double 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] = 1.;
+ dm[2] = 0.;
+ }
+
+ 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];
+
+ for (int n = 0; n < out->nb_samples; n++) {
+ double fa[3], fm[3], sa[3], sm[3];
+ double detect, gain, v;
+
+ detect = get_svf(src[n], dm, da, state);
+ gain = get_gain(ctx, detect, sample_rate, ch);
+
+ {
+ double k = 1. / sqfactor;
+
+ sa[0] = 1. / (1. + sg * (sg + k));
+ sa[1] = sg * sa[0];
+ sa[2] = sg * sa[1];
+
+ sm[0] = 0.;
+ sm[1] = 0.;
+ sm[2] = 1.;
+ }
+
+ gain = get_svf(gain, sm, sa, &state[5]);
+ gain = mode ? av_clipd(gain, 1., irange) : av_clipd(gain, range, 1.);
+
+ {
+ double 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.;
+ }
+
+ v = get_svf(src[n], fm, fa, &state[3]);
+ dst[n] = ctx->is_disabled ? src[n] : v;
+ }
+ }
+
+ return 0;
+}
+
+static int filter_frame(AVFilterLink *inlink, AVFrame *in)
+{
+ AVFilterContext *ctx = inlink->dst;
+ AudioDynamicEqualizerContext *s = ctx->priv;
+ AVFilterLink *outlink = ctx->outputs[0];
+ ThreadData td;
+ AVFrame *out;
+
+ if (av_frame_is_writable(in)) {
+ out = in;
+ } else {
+ out = ff_get_audio_buffer(outlink, in->nb_samples);
+ if (!out) {
+ av_frame_free(&in);
+ return AVERROR(ENOMEM);
+ }
+ av_frame_copy_props(out, in);
+ }
+
+ s->attack_coeff = FFMIN(1., 1. / (s->attack * in->sample_rate / 4000.));
+ s->release_coeff = FFMIN(1., 1. / (s->release * in->sample_rate / 4000.));
+ s->knee_sqrt = sqrt(s->knee);
+ s->lin_knee_stop = s->lin_threshold * s->knee_sqrt;
+ s->lin_knee_start = s->lin_threshold / s->knee_sqrt;
+ s->knee_start = log(s->lin_knee_start);
+ s->knee_stop = log(s->lin_knee_stop);
+ s->threshold = log(s->lin_threshold);
+ s->compressed_knee_start = (s->knee_start - s->threshold) / s->ratio + s->threshold;
+
+ td.in = in;
+ td.out = out;
+ ff_filter_execute(ctx, filter_channels, &td, NULL,
+ FFMIN(outlink->channels, ff_filter_get_nb_threads(ctx)));
+
+ if (out != in)
+ av_frame_free(&in);
+ return ff_filter_frame(outlink, out);
+}
+
+static av_cold void uninit(AVFilterContext *ctx)
+{
+ AudioDynamicEqualizerContext *s = ctx->priv;
+
+ av_frame_free(&s->state);
+}
+
+#define OFFSET(x) offsetof(AudioDynamicEqualizerContext, x)
+#define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
+
+static const AVOption adynamicequalizer_options[] = {
+ { "threshold", "set detection threshold", OFFSET(lin_threshold), AV_OPT_TYPE_DOUBLE, {.dbl=0}, 0, 50, FLAGS },
+ { "dfrequency", "set detection frequency", OFFSET(dfrequency), AV_OPT_TYPE_DOUBLE, {.dbl=1000}, 2, 1000000, FLAGS },
+ { "dqfactor", "set detection Q factor", OFFSET(dqfactor), AV_OPT_TYPE_DOUBLE, {.dbl=1}, 0.001, 1000, FLAGS },
+ { "sfrequency", "set smooth frequency", OFFSET(sfrequency), AV_OPT_TYPE_DOUBLE, {.dbl=100}, 2, 1000000, FLAGS },
+ { "sqfactor", "set smooth Q factor", OFFSET(sqfactor), AV_OPT_TYPE_DOUBLE, {.dbl=1}, 0.001, 1000, FLAGS },
+ { "tfrequency", "set target frequency", OFFSET(tfrequency), AV_OPT_TYPE_DOUBLE, {.dbl=1000}, 2, 1000000, FLAGS },
+ { "tqfactor", "set target Q factor", OFFSET(tqfactor), AV_OPT_TYPE_DOUBLE, {.dbl=1}, 0.001, 1000, FLAGS },
+ { "attack", "set attack", OFFSET(attack), AV_OPT_TYPE_DOUBLE, {.dbl=10}, 1, 2000, FLAGS },
+ { "release", "set release", OFFSET(release), AV_OPT_TYPE_DOUBLE, {.dbl=80}, 1, 2000, FLAGS },
+ { "knee", "set knee factor", OFFSET(knee), AV_OPT_TYPE_DOUBLE, {.dbl=2}, 1, 8, FLAGS },
+ { "ratio", "set ratio factor", OFFSET(ratio), AV_OPT_TYPE_DOUBLE, {.dbl=1}, 1, 20, FLAGS },
+ { "range", "set max gain", OFFSET(range), AV_OPT_TYPE_DOUBLE, {.dbl=0.06125},0.00000001, 1,FLAGS },
+ { "mode", "set mode", OFFSET(mode), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, FLAGS, "mode" },
+ { "cut", 0, 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, FLAGS, "mode" },
+ { "boost", 0, 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, FLAGS, "mode" },
+ { NULL }
+};
+
+AVFILTER_DEFINE_CLASS(adynamicequalizer);
+
+static const AVFilterPad inputs[] = {
+ {
+ .name = "default",
+ .type = AVMEDIA_TYPE_AUDIO,
+ .filter_frame = filter_frame,
+ .config_props = config_input,
+ },
+};
+
+static const AVFilterPad outputs[] = {
+ {
+ .name = "default",
+ .type = AVMEDIA_TYPE_AUDIO,
+ },
+};
+
+const AVFilter ff_af_adynamicequalizer = {
+ .name = "adynamicequalizer",
+ .description = NULL_IF_CONFIG_SMALL("Apply Dynamic Equalization of input audio."),
+ .priv_size = sizeof(AudioDynamicEqualizerContext),
+ .priv_class = &adynamicequalizer_class,
+ .uninit = uninit,
+ FILTER_INPUTS(inputs),
+ FILTER_OUTPUTS(outputs),
+ FILTER_SINGLE_SAMPLEFMT(AV_SAMPLE_FMT_DBLP),
+ .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL |
+ AVFILTER_FLAG_SLICE_THREADS,
+ .process_command = ff_filter_process_command,
+};
diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
index c5c0e9b28b..7018337c85 100644
--- a/libavfilter/allfilters.c
+++ b/libavfilter/allfilters.c
@@ -37,6 +37,7 @@ extern const AVFilter ff_af_adecorrelate;
extern const AVFilter ff_af_adelay;
extern const AVFilter ff_af_adenorm;
extern const AVFilter ff_af_aderivative;
+extern const AVFilter ff_af_adynamicequalizer;
extern const AVFilter ff_af_adynamicsmooth;
extern const AVFilter ff_af_aecho;
extern const AVFilter ff_af_aemphasis;
--
2.33.0
More information about the ffmpeg-devel
mailing list