libavfilter/af_agate.c

   1 /*
   2  * Copyright (C) 2001-2010 Krzysztof Foltman, Markus Schmidt, Thor Harald Johansen, Damien Zammit
   3  *
   4  * This file is part of FFmpeg.
   5  *
   6  * FFmpeg is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2.1 of the License, or (at your option) any later version.
  10  *
  11  * FFmpeg is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with FFmpeg; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 /**
  22  * @file
  23  * Audio (Sidechain) Gate filter
  24  */
  25
  26 #include "config_components.h"
  27
  28 #include "libavutil/audio_fifo.h"
  29 #include "libavutil/channel_layout.h"
  30 #include "libavutil/opt.h"
  31 #include "avfilter.h"
  32 #include "audio.h"
  33 #include "filters.h"
  34 #include "formats.h"
  35 #include "hermite.h"
  36
  37 typedef struct AudioGateContext {
  38     const AVClass *class;
  39
  40     double level_in;
  41     double level_sc;
  42     double attack;
  43     double release;
  44     double threshold;
  45     double ratio;
  46     double knee;
  47     double makeup;
  48     double range;
  49     int link;
  50     int detection;
  51     int mode;
  52
  53     double thres;
  54     double knee_start;
  55     double knee_stop;
  56     double lin_knee_start;
  57     double lin_knee_stop;
  58     double lin_slope;
  59     double attack_coeff;
  60     double release_coeff;
  61
  62     AVAudioFifo *fifo[2];
  63     int64_t pts;
  64 } AudioGateContext;
  65
  66 #define OFFSET(x) offsetof(AudioGateContext, x)
  67 #define A AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
  68
  69 static const AVOption options[] = {
  70     { "level_in",  "set input level",        OFFSET(level_in),  AV_OPT_TYPE_DOUBLE, {.dbl=1},           0.015625,   64, A },
  71     { "mode",      "set mode",               OFFSET(mode),      AV_OPT_TYPE_INT,    {.i64=0},           0, 1, A, .unit = "mode" },
  72     {   "downward",0,                        0,                 AV_OPT_TYPE_CONST,  {.i64=0},           0, 0, A, .unit = "mode" },
  73     {   "upward",  0,                        0,                 AV_OPT_TYPE_CONST,  {.i64=1},           0, 0, A, .unit = "mode" },
  74     { "range",     "set max gain reduction", OFFSET(range),     AV_OPT_TYPE_DOUBLE, {.dbl=0.06125},     0, 1, A },
  75     { "threshold", "set threshold",          OFFSET(threshold), AV_OPT_TYPE_DOUBLE, {.dbl=0.125},       0, 1, A },
  76     { "ratio",     "set ratio",              OFFSET(ratio),     AV_OPT_TYPE_DOUBLE, {.dbl=2},           1,  9000, A },
  77     { "attack",    "set attack",             OFFSET(attack),    AV_OPT_TYPE_DOUBLE, {.dbl=20},          0.01, 9000, A },
  78     { "release",   "set release",            OFFSET(release),   AV_OPT_TYPE_DOUBLE, {.dbl=250},         0.01, 9000, A },
  79     { "makeup",    "set makeup gain",        OFFSET(makeup),    AV_OPT_TYPE_DOUBLE, {.dbl=1},           1,   64, A },
  80     { "knee",      "set knee",               OFFSET(knee),      AV_OPT_TYPE_DOUBLE, {.dbl=2.828427125}, 1,    8, A },
  81     { "detection", "set detection",          OFFSET(detection), AV_OPT_TYPE_INT,    {.i64=1},           0,    1, A, .unit = "detection" },
  82     {   "peak",    0,                        0,                 AV_OPT_TYPE_CONST,  {.i64=0},           0,    0, A, .unit = "detection" },
  83     {   "rms",     0,                        0,                 AV_OPT_TYPE_CONST,  {.i64=1},           0,    0, A, .unit = "detection" },
  84     { "link",      "set link",               OFFSET(link),      AV_OPT_TYPE_INT,    {.i64=0},           0,    1, A, .unit = "link" },
  85     {   "average", 0,                        0,                 AV_OPT_TYPE_CONST,  {.i64=0},           0,    0, A, .unit = "link" },
  86     {   "maximum", 0,                        0,                 AV_OPT_TYPE_CONST,  {.i64=1},           0,    0, A, .unit = "link" },
  87     { "level_sc",  "set sidechain gain",     OFFSET(level_sc),  AV_OPT_TYPE_DOUBLE, {.dbl=1},           0.015625,   64, A },
  88     { NULL }
  89 };
  90
  91 AVFILTER_DEFINE_CLASS_EXT(agate_sidechaingate, "agate/sidechaingate", options);
  92
  93 static int agate_config_input(AVFilterLink *inlink)
  94 {
  95     AVFilterContext *ctx = inlink->dst;
  96     AudioGateContext *s = ctx->priv;
  97     double lin_threshold = s->threshold;
  98     double lin_knee_sqrt = sqrt(s->knee);
  99
 100     if (s->detection)
 101         lin_threshold *= lin_threshold;
 102
 103     s->attack_coeff  = FFMIN(1., 1. / (s->attack * inlink->sample_rate / 4000.));
 104     s->release_coeff = FFMIN(1., 1. / (s->release * inlink->sample_rate / 4000.));
 105     s->lin_knee_stop = lin_threshold * lin_knee_sqrt;
 106     s->lin_knee_start = lin_threshold / lin_knee_sqrt;
 107     s->thres = log(lin_threshold);
 108     s->knee_start = log(s->lin_knee_start);
 109     s->knee_stop = log(s->lin_knee_stop);
 110
 111     return 0;
 112 }
 113
 114 // A fake infinity value (because real infinity may break some hosts)
 115 #define FAKE_INFINITY (65536.0 * 65536.0)
 116
 117 // Check for infinity (with appropriate-ish tolerance)
 118 #define IS_FAKE_INFINITY(value) (fabs(value-FAKE_INFINITY) < 1.0)
 119
 120 static double output_gain(double lin_slope, double ratio, double thres,
 121                           double knee, double knee_start, double knee_stop,
 122                           double range, int mode)
 123 {
 124     double slope = log(lin_slope);
 125     double tratio = ratio;
 126     double gain = 0.;
 127     double delta = 0.;
 128
 129     if (IS_FAKE_INFINITY(ratio))
 130         tratio = 1000.;
 131     gain = (slope - thres) * tratio + thres;
 132     delta = tratio;
 133
 134     if (mode) {
 135         if (knee > 1. && slope < knee_stop)
 136             gain = hermite_interpolation(slope, knee_stop, knee_start, ((knee_stop - thres) * tratio  + thres), knee_start, delta, 1.);
 137     } else {
 138         if (knee > 1. && slope > knee_start)
 139             gain = hermite_interpolation(slope, knee_start, knee_stop, ((knee_start - thres) * tratio  + thres), knee_stop, delta, 1.);
 140     }
 141     return FFMAX(range, exp(gain - slope));
 142 }
 143
 144 static void gate(AudioGateContext *s,
 145                  const double *src, double *dst, const double *scsrc,
 146                  int nb_samples, double level_in, double level_sc,
 147                  AVFilterLink *inlink, AVFilterLink *sclink)
 148 {
 149     AVFilterContext *ctx = inlink->dst;
 150     const double makeup = s->makeup;
 151     const double attack_coeff = s->attack_coeff;
 152     const double release_coeff = s->release_coeff;
 153     int n, c;
 154
 155     for (n = 0; n < nb_samples; n++, src += inlink->ch_layout.nb_channels, dst += inlink->ch_layout.nb_channels, scsrc += sclink->ch_layout.nb_channels) {
 156         double abs_sample = fabs(scsrc[0] * level_sc), gain = 1.0;
 157         double factor;
 158         int detected;
 159
 160         if (s->link == 1) {
 161             for (c = 1; c < sclink->ch_layout.nb_channels; c++)
 162                 abs_sample = FFMAX(fabs(scsrc[c] * level_sc), abs_sample);
 163         } else {
 164             for (c = 1; c < sclink->ch_layout.nb_channels; c++)
 165                 abs_sample += fabs(scsrc[c] * level_sc);
 166
 167             abs_sample /= sclink->ch_layout.nb_channels;
 168         }
 169
 170         if (s->detection)
 171             abs_sample *= abs_sample;
 172
 173         s->lin_slope += (abs_sample - s->lin_slope) * (abs_sample > s->lin_slope ? attack_coeff : release_coeff);
 174
 175         if (s->mode)
 176             detected = s->lin_slope > s->lin_knee_start;
 177         else
 178             detected = s->lin_slope < s->lin_knee_stop;
 179
 180         if (s->lin_slope > 0.0 && detected)
 181             gain = output_gain(s->lin_slope, s->ratio, s->thres,
 182                                s->knee, s->knee_start, s->knee_stop,
 183                                s->range, s->mode);
 184
 185         factor = ctx->is_disabled ? 1.f : level_in * gain * makeup;
 186         for (c = 0; c < inlink->ch_layout.nb_channels; c++)
 187             dst[c] = src[c] * factor;
 188     }
 189 }
 190
 191 #if CONFIG_AGATE_FILTER
 192
 193 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
 194 {
 195     const double *src = (const double *)in->data[0];
 196     AVFilterContext *ctx = inlink->dst;
 197     AVFilterLink *outlink = ctx->outputs[0];
 198     AudioGateContext *s = ctx->priv;
 199     AVFrame *out;
 200     double *dst;
 201
 202     if (av_frame_is_writable(in)) {
 203         out = in;
 204     } else {
 205         out = ff_get_audio_buffer(outlink, in->nb_samples);
 206         if (!out) {
 207             av_frame_free(&in);
 208             return AVERROR(ENOMEM);
 209         }
 210         av_frame_copy_props(out, in);
 211     }
 212     dst = (double *)out->data[0];
 213
 214     gate(s, src, dst, src, in->nb_samples,
 215          s->level_in, s->level_in, inlink, inlink);
 216
 217     if (out != in)
 218         av_frame_free(&in);
 219     return ff_filter_frame(outlink, out);
 220 }
 221
 222 static const AVFilterPad inputs[] = {
 223     {
 224         .name         = "default",
 225         .type         = AVMEDIA_TYPE_AUDIO,
 226         .filter_frame = filter_frame,
 227         .config_props = agate_config_input,
 228     },
 229 };
 230
 231 const FFFilter ff_af_agate = {
 232     .p.name         = "agate",
 233     .p.description  = NULL_IF_CONFIG_SMALL("Audio gate."),
 234     .p.priv_class   = &agate_sidechaingate_class,
 235     .p.flags        = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC,
 236     .priv_size      = sizeof(AudioGateContext),
 237     FILTER_INPUTS(inputs),
 238     FILTER_OUTPUTS(ff_audio_default_filterpad),
 239     FILTER_SINGLE_SAMPLEFMT(AV_SAMPLE_FMT_DBL),
 240     .process_command = ff_filter_process_command,
 241 };
 242
 243 #endif /* CONFIG_AGATE_FILTER */
 244
 245 #if CONFIG_SIDECHAINGATE_FILTER
 246
 247 static int activate(AVFilterContext *ctx)
 248 {
 249     AudioGateContext *s = ctx->priv;
 250     AVFrame *out = NULL, *in[2] = { NULL };
 251     int ret, i, nb_samples;
 252     double *dst;
 253
 254     FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx);
 255     if ((ret = ff_inlink_consume_frame(ctx->inputs[0], &in[0])) > 0) {
 256         av_audio_fifo_write(s->fifo[0], (void **)in[0]->extended_data,
 257                             in[0]->nb_samples);
 258         av_frame_free(&in[0]);
 259     }
 260     if (ret < 0)
 261         return ret;
 262     if ((ret = ff_inlink_consume_frame(ctx->inputs[1], &in[1])) > 0) {
 263         av_audio_fifo_write(s->fifo[1], (void **)in[1]->extended_data,
 264                             in[1]->nb_samples);
 265         av_frame_free(&in[1]);
 266     }
 267     if (ret < 0)
 268         return ret;
 269
 270     nb_samples = FFMIN(av_audio_fifo_size(s->fifo[0]), av_audio_fifo_size(s->fifo[1]));
 271     if (nb_samples) {
 272         out = ff_get_audio_buffer(ctx->outputs[0], nb_samples);
 273         if (!out)
 274             return AVERROR(ENOMEM);
 275         for (i = 0; i < 2; i++) {
 276             in[i] = ff_get_audio_buffer(ctx->inputs[i], nb_samples);
 277             if (!in[i]) {
 278                 av_frame_free(&in[0]);
 279                 av_frame_free(&in[1]);
 280                 av_frame_free(&out);
 281                 return AVERROR(ENOMEM);
 282             }
 283             av_audio_fifo_read(s->fifo[i], (void **)in[i]->data, nb_samples);
 284         }
 285
 286         dst = (double *)out->data[0];
 287         out->pts = s->pts;
 288         s->pts += av_rescale_q(nb_samples, (AVRational){1, ctx->outputs[0]->sample_rate}, ctx->outputs[0]->time_base);
 289
 290         gate(s, (double *)in[0]->data[0], dst,
 291              (double *)in[1]->data[0], nb_samples,
 292              s->level_in, s->level_sc,
 293              ctx->inputs[0], ctx->inputs[1]);
 294
 295         av_frame_free(&in[0]);
 296         av_frame_free(&in[1]);
 297
 298         ret = ff_filter_frame(ctx->outputs[0], out);
 299         if (ret < 0)
 300             return ret;
 301     }
 302     FF_FILTER_FORWARD_STATUS(ctx->inputs[0], ctx->outputs[0]);
 303     FF_FILTER_FORWARD_STATUS(ctx->inputs[1], ctx->outputs[0]);
 304     if (ff_outlink_frame_wanted(ctx->outputs[0])) {
 305         if (!av_audio_fifo_size(s->fifo[0]))
 306             ff_inlink_request_frame(ctx->inputs[0]);
 307         if (!av_audio_fifo_size(s->fifo[1]))
 308             ff_inlink_request_frame(ctx->inputs[1]);
 309     }
 310     return 0;
 311 }
 312
 313 static int scquery_formats(const AVFilterContext *ctx,
 314                            AVFilterFormatsConfig **cfg_in,
 315                            AVFilterFormatsConfig **cfg_out)
 316 {
 317     static const enum AVSampleFormat sample_fmts[] = {
 318         AV_SAMPLE_FMT_DBL,
 319         AV_SAMPLE_FMT_NONE
 320     };
 321     int ret;
 322
 323     /* Generic code will link the channel properties of the main input and the
 324      * output; it won't touch the second input as its channel_layouts is already
 325      * set. */
 326     ret = ff_channel_layouts_ref(ff_all_channel_counts(),
 327                                  &cfg_in[1]->channel_layouts);
 328     if (ret < 0)
 329         return ret;
 330
 331     if ((ret = ff_set_common_formats_from_list2(ctx, cfg_in, cfg_out, sample_fmts)) < 0)
 332         return ret;
 333
 334     return 0;
 335 }
 336
 337 static int scconfig_output(AVFilterLink *outlink)
 338 {
 339     AVFilterContext *ctx = outlink->src;
 340     AudioGateContext *s = ctx->priv;
 341
 342     outlink->time_base   = ctx->inputs[0]->time_base;
 343
 344     s->fifo[0] = av_audio_fifo_alloc(ctx->inputs[0]->format, ctx->inputs[0]->ch_layout.nb_channels, 1024);
 345     s->fifo[1] = av_audio_fifo_alloc(ctx->inputs[1]->format, ctx->inputs[1]->ch_layout.nb_channels, 1024);
 346     if (!s->fifo[0] || !s->fifo[1])
 347         return AVERROR(ENOMEM);
 348
 349
 350     agate_config_input(ctx->inputs[0]);
 351
 352     return 0;
 353 }
 354
 355 static av_cold void uninit(AVFilterContext *ctx)
 356 {
 357     AudioGateContext *s = ctx->priv;
 358
 359     av_audio_fifo_free(s->fifo[0]);
 360     av_audio_fifo_free(s->fifo[1]);
 361 }
 362
 363 static const AVFilterPad sidechaingate_inputs[] = {
 364     {
 365         .name           = "main",
 366         .type           = AVMEDIA_TYPE_AUDIO,
 367     },{
 368         .name           = "sidechain",
 369         .type           = AVMEDIA_TYPE_AUDIO,
 370     },
 371 };
 372
 373 static const AVFilterPad sidechaingate_outputs[] = {
 374     {
 375         .name          = "default",
 376         .type          = AVMEDIA_TYPE_AUDIO,
 377         .config_props  = scconfig_output,
 378     },
 379 };
 380
 381 const FFFilter ff_af_sidechaingate = {
 382     .p.name         = "sidechaingate",
 383     .p.description  = NULL_IF_CONFIG_SMALL("Audio sidechain gate."),
 384     .p.priv_class   = &agate_sidechaingate_class,
 385     .p.flags        = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
 386     .priv_size      = sizeof(AudioGateContext),
 387     .activate       = activate,
 388     .uninit         = uninit,
 389     FILTER_INPUTS(sidechaingate_inputs),
 390     FILTER_OUTPUTS(sidechaingate_outputs),
 391     FILTER_QUERY_FUNC2(scquery_formats),
 392     .process_command = ff_filter_process_command,
 393 };
 394 #endif  /* CONFIG_SIDECHAINGATE_FILTER */