git.sesse.net Git - ffmpeg/blob - libavfilter/af_silenceremove.c

   1 /*
   2  * Copyright (c) 2001 Heikki Leinonen
   3  * Copyright (c) 2001 Chris Bagwell
   4  * Copyright (c) 2003 Donnie Smith
   5  * Copyright (c) 2014 Paul B Mahol
   6  *
   7  * This file is part of FFmpeg.
   8  *
   9  * FFmpeg is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * FFmpeg is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with FFmpeg; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23
  24 #include <float.h> /* DBL_MAX */
  25
  26 #include "libavutil/opt.h"
  27 #include "libavutil/timestamp.h"
  28 #include "audio.h"
  29 #include "formats.h"
  30 #include "avfilter.h"
  31 #include "internal.h"
  32
  33 enum SilenceMode {
  34     SILENCE_TRIM,
  35     SILENCE_TRIM_FLUSH,
  36     SILENCE_COPY,
  37     SILENCE_COPY_FLUSH,
  38     SILENCE_STOP
  39 };
  40
  41 typedef struct SilenceRemoveContext {
  42     const AVClass *class;
  43
  44     enum SilenceMode mode;
  45
  46     int start_periods;
  47     int64_t start_duration;
  48     double start_threshold;
  49
  50     int stop_periods;
  51     int64_t stop_duration;
  52     double stop_threshold;
  53
  54     double *start_holdoff;
  55     size_t start_holdoff_offset;
  56     size_t start_holdoff_end;
  57     int    start_found_periods;
  58
  59     double *stop_holdoff;
  60     size_t stop_holdoff_offset;
  61     size_t stop_holdoff_end;
  62     int    stop_found_periods;
  63
  64     double window_ratio;
  65     double *window;
  66     double *window_current;
  67     double *window_end;
  68     int window_size;
  69     double sum;
  70
  71     int leave_silence;
  72     int restart;
  73     int64_t next_pts;
  74
  75     int detection;
  76     void (*update)(struct SilenceRemoveContext *s, double sample);
  77     double(*compute)(struct SilenceRemoveContext *s, double sample);
  78 } SilenceRemoveContext;
  79
  80 #define OFFSET(x) offsetof(SilenceRemoveContext, x)
  81 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_AUDIO_PARAM
  82 static const AVOption silenceremove_options[] = {
  83     { "start_periods",   NULL, OFFSET(start_periods),   AV_OPT_TYPE_INT,      {.i64=0},     0,    9000, FLAGS },
  84     { "start_duration",  NULL, OFFSET(start_duration),  AV_OPT_TYPE_DURATION, {.i64=0},     0,    9000, FLAGS },
  85     { "start_threshold", NULL, OFFSET(start_threshold), AV_OPT_TYPE_DOUBLE,   {.dbl=0},     0, DBL_MAX, FLAGS },
  86     { "stop_periods",    NULL, OFFSET(stop_periods),    AV_OPT_TYPE_INT,      {.i64=0}, -9000,    9000, FLAGS },
  87     { "stop_duration",   NULL, OFFSET(stop_duration),   AV_OPT_TYPE_DURATION, {.i64=0},     0,    9000, FLAGS },
  88     { "stop_threshold",  NULL, OFFSET(stop_threshold),  AV_OPT_TYPE_DOUBLE,   {.dbl=0},     0, DBL_MAX, FLAGS },
  89     { "leave_silence",   NULL, OFFSET(leave_silence),   AV_OPT_TYPE_BOOL,     {.i64=0},     0,       1, FLAGS },
  90     { "detection",       NULL, OFFSET(detection),       AV_OPT_TYPE_INT,      {.i64=1},     0,       1, FLAGS, "detection" },
  91     {   "peak",          0,    0,                       AV_OPT_TYPE_CONST,    {.i64=0},     0,       0, FLAGS, "detection" },
  92     {   "rms",           0,    0,                       AV_OPT_TYPE_CONST,    {.i64=1},     0,       0, FLAGS, "detection" },
  93     { "window",          NULL, OFFSET(window_ratio),    AV_OPT_TYPE_DOUBLE,   {.dbl=0.02},  0,      10, FLAGS },
  94     { NULL }
  95 };
  96
  97 AVFILTER_DEFINE_CLASS(silenceremove);
  98
  99 static double compute_peak(SilenceRemoveContext *s, double sample)
 100 {
 101     double new_sum;
 102
 103     new_sum  = s->sum;
 104     new_sum -= *s->window_current;
 105     new_sum += fabs(sample);
 106
 107     return new_sum / s->window_size;
 108 }
 109
 110 static void update_peak(SilenceRemoveContext *s, double sample)
 111 {
 112     s->sum -= *s->window_current;
 113     *s->window_current = fabs(sample);
 114     s->sum += *s->window_current;
 115
 116     s->window_current++;
 117     if (s->window_current >= s->window_end)
 118         s->window_current = s->window;
 119 }
 120
 121 static double compute_rms(SilenceRemoveContext *s, double sample)
 122 {
 123     double new_sum;
 124
 125     new_sum  = s->sum;
 126     new_sum -= *s->window_current;
 127     new_sum += sample * sample;
 128
 129     return sqrt(new_sum / s->window_size);
 130 }
 131
 132 static void update_rms(SilenceRemoveContext *s, double sample)
 133 {
 134     s->sum -= *s->window_current;
 135     *s->window_current = sample * sample;
 136     s->sum += *s->window_current;
 137
 138     s->window_current++;
 139     if (s->window_current >= s->window_end)
 140         s->window_current = s->window;
 141 }
 142
 143 static av_cold int init(AVFilterContext *ctx)
 144 {
 145     SilenceRemoveContext *s = ctx->priv;
 146
 147     if (s->stop_periods < 0) {
 148         s->stop_periods = -s->stop_periods;
 149         s->restart = 1;
 150     }
 151
 152     switch (s->detection) {
 153     case 0:
 154         s->update = update_peak;
 155         s->compute = compute_peak;
 156         break;
 157     case 1:
 158         s->update = update_rms;
 159         s->compute = compute_rms;
 160         break;
 161     };
 162
 163     return 0;
 164 }
 165
 166 static void clear_window(SilenceRemoveContext *s)
 167 {
 168     memset(s->window, 0, s->window_size * sizeof(*s->window));
 169
 170     s->window_current = s->window;
 171     s->window_end = s->window + s->window_size;
 172     s->sum = 0;
 173 }
 174
 175 static int config_input(AVFilterLink *inlink)
 176 {
 177     AVFilterContext *ctx = inlink->dst;
 178     SilenceRemoveContext *s = ctx->priv;
 179
 180     s->window_size = FFMAX((inlink->sample_rate * s->window_ratio), 1) * inlink->channels;
 181     s->window = av_malloc_array(s->window_size, sizeof(*s->window));
 182     if (!s->window)
 183         return AVERROR(ENOMEM);
 184
 185     clear_window(s);
 186
 187     s->start_duration = av_rescale(s->start_duration, inlink->sample_rate,
 188                                    AV_TIME_BASE);
 189     if (s->start_duration < 0) {
 190         av_log(ctx, AV_LOG_WARNING, "start duration must be non-negative\n");
 191         s->start_duration = -s->start_duration;
 192     }
 193
 194     s->stop_duration  = av_rescale(s->stop_duration, inlink->sample_rate,
 195                                    AV_TIME_BASE);
 196     if (s->stop_duration < 0) {
 197         av_log(ctx, AV_LOG_WARNING, "stop duration must be non-negative\n");
 198         s->stop_duration = -s->stop_duration;
 199     }
 200
 201     s->start_holdoff = av_malloc_array(FFMAX(s->start_duration, 1),
 202                                        sizeof(*s->start_holdoff) *
 203                                        inlink->channels);
 204     if (!s->start_holdoff)
 205         return AVERROR(ENOMEM);
 206
 207     s->start_holdoff_offset = 0;
 208     s->start_holdoff_end    = 0;
 209     s->start_found_periods  = 0;
 210
 211     s->stop_holdoff = av_malloc_array(FFMAX(s->stop_duration, 1),
 212                                       sizeof(*s->stop_holdoff) *
 213                                       inlink->channels);
 214     if (!s->stop_holdoff)
 215         return AVERROR(ENOMEM);
 216
 217     s->stop_holdoff_offset = 0;
 218     s->stop_holdoff_end    = 0;
 219     s->stop_found_periods  = 0;
 220
 221     if (s->start_periods)
 222         s->mode = SILENCE_TRIM;
 223     else
 224         s->mode = SILENCE_COPY;
 225
 226     return 0;
 227 }
 228
 229 static void flush(SilenceRemoveContext *s,
 230                   AVFrame *out, AVFilterLink *outlink,
 231                   int *nb_samples_written, int *ret)
 232 {
 233     if (*nb_samples_written) {
 234         out->nb_samples = *nb_samples_written / outlink->channels;
 235
 236         out->pts = s->next_pts;
 237         s->next_pts += av_rescale_q(out->nb_samples,
 238                                     (AVRational){1, outlink->sample_rate},
 239                                     outlink->time_base);
 240
 241         *ret = ff_filter_frame(outlink, out);
 242         *nb_samples_written = 0;
 243     } else {
 244         av_frame_free(&out);
 245     }
 246 }
 247
 248 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
 249 {
 250     AVFilterContext *ctx = inlink->dst;
 251     AVFilterLink *outlink = ctx->outputs[0];
 252     SilenceRemoveContext *s = ctx->priv;
 253     int i, j, threshold, ret = 0;
 254     int nbs, nb_samples_read, nb_samples_written;
 255     double *obuf, *ibuf = (double *)in->data[0];
 256     AVFrame *out;
 257
 258     nb_samples_read = nb_samples_written = 0;
 259
 260     switch (s->mode) {
 261     case SILENCE_TRIM:
 262 silence_trim:
 263         nbs = in->nb_samples - nb_samples_read / inlink->channels;
 264         if (!nbs)
 265             break;
 266
 267         for (i = 0; i < nbs; i++) {
 268             threshold = 0;
 269             for (j = 0; j < inlink->channels; j++) {
 270                 threshold |= s->compute(s, ibuf[j]) > s->start_threshold;
 271             }
 272
 273             if (threshold) {
 274                 for (j = 0; j < inlink->channels; j++) {
 275                     s->update(s, *ibuf);
 276                     s->start_holdoff[s->start_holdoff_end++] = *ibuf++;
 277                 }
 278                 nb_samples_read += inlink->channels;
 279
 280                 if (s->start_holdoff_end >= s->start_duration * inlink->channels) {
 281                     if (++s->start_found_periods >= s->start_periods) {
 282                         s->mode = SILENCE_TRIM_FLUSH;
 283                         goto silence_trim_flush;
 284                     }
 285
 286                     s->start_holdoff_offset = 0;
 287                     s->start_holdoff_end = 0;
 288                 }
 289             } else {
 290                 s->start_holdoff_end = 0;
 291
 292                 for (j = 0; j < inlink->channels; j++)
 293                     s->update(s, ibuf[j]);
 294
 295                 ibuf += inlink->channels;
 296                 nb_samples_read += inlink->channels;
 297             }
 298         }
 299         break;
 300
 301     case SILENCE_TRIM_FLUSH:
 302 silence_trim_flush:
 303         nbs  = s->start_holdoff_end - s->start_holdoff_offset;
 304         nbs -= nbs % inlink->channels;
 305         if (!nbs)
 306             break;
 307
 308         out = ff_get_audio_buffer(inlink, nbs / inlink->channels);
 309         if (!out) {
 310             av_frame_free(&in);
 311             return AVERROR(ENOMEM);
 312         }
 313
 314         memcpy(out->data[0], &s->start_holdoff[s->start_holdoff_offset],
 315                nbs * sizeof(double));
 316
 317         out->pts = s->next_pts;
 318         s->next_pts += av_rescale_q(out->nb_samples,
 319                                     (AVRational){1, outlink->sample_rate},
 320                                     outlink->time_base);
 321
 322         s->start_holdoff_offset += nbs;
 323
 324         ret = ff_filter_frame(outlink, out);
 325
 326         if (s->start_holdoff_offset == s->start_holdoff_end) {
 327             s->start_holdoff_offset = 0;
 328             s->start_holdoff_end = 0;
 329             s->mode = SILENCE_COPY;
 330             goto silence_copy;
 331         }
 332         break;
 333
 334     case SILENCE_COPY:
 335 silence_copy:
 336         nbs = in->nb_samples - nb_samples_read / inlink->channels;
 337         if (!nbs)
 338             break;
 339
 340         out = ff_get_audio_buffer(inlink, nbs);
 341         if (!out) {
 342             av_frame_free(&in);
 343             return AVERROR(ENOMEM);
 344         }
 345         obuf = (double *)out->data[0];
 346
 347         if (s->stop_periods) {
 348             for (i = 0; i < nbs; i++) {
 349                 threshold = 1;
 350                 for (j = 0; j < inlink->channels; j++)
 351                     threshold &= s->compute(s, ibuf[j]) > s->stop_threshold;
 352
 353                 if (threshold && s->stop_holdoff_end && !s->leave_silence) {
 354                     s->mode = SILENCE_COPY_FLUSH;
 355                     flush(s, out, outlink, &nb_samples_written, &ret);
 356                     goto silence_copy_flush;
 357                 } else if (threshold) {
 358                     for (j = 0; j < inlink->channels; j++) {
 359                         s->update(s, *ibuf);
 360                         *obuf++ = *ibuf++;
 361                     }
 362                     nb_samples_read    += inlink->channels;
 363                     nb_samples_written += inlink->channels;
 364                 } else if (!threshold) {
 365                     for (j = 0; j < inlink->channels; j++) {
 366                         s->update(s, *ibuf);
 367                         if (s->leave_silence) {
 368                             *obuf++ = *ibuf;
 369                             nb_samples_written++;
 370                         }
 371
 372                         s->stop_holdoff[s->stop_holdoff_end++] = *ibuf++;
 373                     }
 374                     nb_samples_read += inlink->channels;
 375
 376                     if (s->stop_holdoff_end >= s->stop_duration * inlink->channels) {
 377                         if (++s->stop_found_periods >= s->stop_periods) {
 378                             s->stop_holdoff_offset = 0;
 379                             s->stop_holdoff_end = 0;
 380
 381                             if (!s->restart) {
 382                                 s->mode = SILENCE_STOP;
 383                                 flush(s, out, outlink, &nb_samples_written, &ret);
 384                                 goto silence_stop;
 385                             } else {
 386                                 s->stop_found_periods = 0;
 387                                 s->start_found_periods = 0;
 388                                 s->start_holdoff_offset = 0;
 389                                 s->start_holdoff_end = 0;
 390                                 clear_window(s);
 391                                 s->mode = SILENCE_TRIM;
 392                                 flush(s, out, outlink, &nb_samples_written, &ret);
 393                                 goto silence_trim;
 394                             }
 395                         }
 396                         s->mode = SILENCE_COPY_FLUSH;
 397                         flush(s, out, outlink, &nb_samples_written, &ret);
 398                         goto silence_copy_flush;
 399                     }
 400                 }
 401             }
 402             flush(s, out, outlink, &nb_samples_written, &ret);
 403         } else {
 404             memcpy(obuf, ibuf, sizeof(double) * nbs * inlink->channels);
 405
 406             out->pts = s->next_pts;
 407             s->next_pts += av_rescale_q(out->nb_samples,
 408                                         (AVRational){1, outlink->sample_rate},
 409                                         outlink->time_base);
 410
 411             ret = ff_filter_frame(outlink, out);
 412         }
 413         break;
 414
 415     case SILENCE_COPY_FLUSH:
 416 silence_copy_flush:
 417         nbs  = s->stop_holdoff_end - s->stop_holdoff_offset;
 418         nbs -= nbs % inlink->channels;
 419         if (!nbs)
 420             break;
 421
 422         out = ff_get_audio_buffer(inlink, nbs / inlink->channels);
 423         if (!out) {
 424             av_frame_free(&in);
 425             return AVERROR(ENOMEM);
 426         }
 427
 428         memcpy(out->data[0], &s->stop_holdoff[s->stop_holdoff_offset],
 429                nbs * sizeof(double));
 430         s->stop_holdoff_offset += nbs;
 431
 432         out->pts = s->next_pts;
 433         s->next_pts += av_rescale_q(out->nb_samples,
 434                                     (AVRational){1, outlink->sample_rate},
 435                                     outlink->time_base);
 436
 437         ret = ff_filter_frame(outlink, out);
 438
 439         if (s->stop_holdoff_offset == s->stop_holdoff_end) {
 440             s->stop_holdoff_offset = 0;
 441             s->stop_holdoff_end = 0;
 442             s->mode = SILENCE_COPY;
 443             goto silence_copy;
 444         }
 445         break;
 446     case SILENCE_STOP:
 447 silence_stop:
 448         break;
 449     }
 450
 451     av_frame_free(&in);
 452
 453     return ret;
 454 }
 455
 456 static int request_frame(AVFilterLink *outlink)
 457 {
 458     AVFilterContext *ctx = outlink->src;
 459     SilenceRemoveContext *s = ctx->priv;
 460     int ret;
 461
 462     ret = ff_request_frame(ctx->inputs[0]);
 463     if (ret == AVERROR_EOF && (s->mode == SILENCE_COPY_FLUSH ||
 464                                s->mode == SILENCE_COPY)) {
 465         int nbs = s->stop_holdoff_end - s->stop_holdoff_offset;
 466         if (nbs) {
 467             AVFrame *frame;
 468
 469             frame = ff_get_audio_buffer(outlink, nbs / outlink->channels);
 470             if (!frame)
 471                 return AVERROR(ENOMEM);
 472
 473             memcpy(frame->data[0], &s->stop_holdoff[s->stop_holdoff_offset],
 474                    nbs * sizeof(double));
 475
 476             frame->pts = s->next_pts;
 477             s->next_pts += av_rescale_q(frame->nb_samples,
 478                                         (AVRational){1, outlink->sample_rate},
 479                                         outlink->time_base);
 480
 481             ret = ff_filter_frame(outlink, frame);
 482         }
 483         s->mode = SILENCE_STOP;
 484     }
 485     return ret;
 486 }
 487
 488 static int query_formats(AVFilterContext *ctx)
 489 {
 490     AVFilterFormats *formats = NULL;
 491     AVFilterChannelLayouts *layouts = NULL;
 492     static const enum AVSampleFormat sample_fmts[] = {
 493         AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_NONE
 494     };
 495     int ret;
 496
 497     layouts = ff_all_channel_counts();
 498     if (!layouts)
 499         return AVERROR(ENOMEM);
 500     ret = ff_set_common_channel_layouts(ctx, layouts);
 501     if (ret < 0)
 502         return ret;
 503
 504     formats = ff_make_format_list(sample_fmts);
 505     if (!formats)
 506         return AVERROR(ENOMEM);
 507     ret = ff_set_common_formats(ctx, formats);
 508     if (ret < 0)
 509         return ret;
 510
 511     formats = ff_all_samplerates();
 512     if (!formats)
 513         return AVERROR(ENOMEM);
 514     return ff_set_common_samplerates(ctx, formats);
 515 }
 516
 517 static av_cold void uninit(AVFilterContext *ctx)
 518 {
 519     SilenceRemoveContext *s = ctx->priv;
 520
 521     av_freep(&s->start_holdoff);
 522     av_freep(&s->stop_holdoff);
 523     av_freep(&s->window);
 524 }
 525
 526 static const AVFilterPad silenceremove_inputs[] = {
 527     {
 528         .name         = "default",
 529         .type         = AVMEDIA_TYPE_AUDIO,
 530         .config_props = config_input,
 531         .filter_frame = filter_frame,
 532     },
 533     { NULL }
 534 };
 535
 536 static const AVFilterPad silenceremove_outputs[] = {
 537     {
 538         .name          = "default",
 539         .type          = AVMEDIA_TYPE_AUDIO,
 540         .request_frame = request_frame,
 541     },
 542     { NULL }
 543 };
 544
 545 AVFilter ff_af_silenceremove = {
 546     .name          = "silenceremove",
 547     .description   = NULL_IF_CONFIG_SMALL("Remove silence."),
 548     .priv_size     = sizeof(SilenceRemoveContext),
 549     .priv_class    = &silenceremove_class,
 550     .init          = init,
 551     .uninit        = uninit,
 552     .query_formats = query_formats,
 553     .inputs        = silenceremove_inputs,
 554     .outputs       = silenceremove_outputs,
 555 };