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     s->stop_duration  = av_rescale(s->stop_duration, inlink->sample_rate,
 190                                    AV_TIME_BASE);
 191
 192     s->start_holdoff = av_malloc_array(FFMAX(s->start_duration, 1),
 193                                        sizeof(*s->start_holdoff) *
 194                                        inlink->channels);
 195     if (!s->start_holdoff)
 196         return AVERROR(ENOMEM);
 197
 198     s->start_holdoff_offset = 0;
 199     s->start_holdoff_end    = 0;
 200     s->start_found_periods  = 0;
 201
 202     s->stop_holdoff = av_malloc_array(FFMAX(s->stop_duration, 1),
 203                                       sizeof(*s->stop_holdoff) *
 204                                       inlink->channels);
 205     if (!s->stop_holdoff)
 206         return AVERROR(ENOMEM);
 207
 208     s->stop_holdoff_offset = 0;
 209     s->stop_holdoff_end    = 0;
 210     s->stop_found_periods  = 0;
 211
 212     if (s->start_periods)
 213         s->mode = SILENCE_TRIM;
 214     else
 215         s->mode = SILENCE_COPY;
 216
 217     return 0;
 218 }
 219
 220 static void flush(SilenceRemoveContext *s,
 221                   AVFrame *out, AVFilterLink *outlink,
 222                   int *nb_samples_written, int *ret)
 223 {
 224     if (*nb_samples_written) {
 225         out->nb_samples = *nb_samples_written / outlink->channels;
 226
 227         out->pts = s->next_pts;
 228         s->next_pts += av_rescale_q(out->nb_samples,
 229                                     (AVRational){1, outlink->sample_rate},
 230                                     outlink->time_base);
 231
 232         *ret = ff_filter_frame(outlink, out);
 233         *nb_samples_written = 0;
 234     } else {
 235         av_frame_free(&out);
 236     }
 237 }
 238
 239 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
 240 {
 241     AVFilterContext *ctx = inlink->dst;
 242     AVFilterLink *outlink = ctx->outputs[0];
 243     SilenceRemoveContext *s = ctx->priv;
 244     int i, j, threshold, ret = 0;
 245     int nbs, nb_samples_read, nb_samples_written;
 246     double *obuf, *ibuf = (double *)in->data[0];
 247     AVFrame *out;
 248
 249     nb_samples_read = nb_samples_written = 0;
 250
 251     switch (s->mode) {
 252     case SILENCE_TRIM:
 253 silence_trim:
 254         nbs = in->nb_samples - nb_samples_read / inlink->channels;
 255         if (!nbs)
 256             break;
 257
 258         for (i = 0; i < nbs; i++) {
 259             threshold = 0;
 260             for (j = 0; j < inlink->channels; j++) {
 261                 threshold |= s->compute(s, ibuf[j]) > s->start_threshold;
 262             }
 263
 264             if (threshold) {
 265                 for (j = 0; j < inlink->channels; j++) {
 266                     s->update(s, *ibuf);
 267                     s->start_holdoff[s->start_holdoff_end++] = *ibuf++;
 268                 }
 269                 nb_samples_read += inlink->channels;
 270
 271                 if (s->start_holdoff_end >= s->start_duration * inlink->channels) {
 272                     if (++s->start_found_periods >= s->start_periods) {
 273                         s->mode = SILENCE_TRIM_FLUSH;
 274                         goto silence_trim_flush;
 275                     }
 276
 277                     s->start_holdoff_offset = 0;
 278                     s->start_holdoff_end = 0;
 279                 }
 280             } else {
 281                 s->start_holdoff_end = 0;
 282
 283                 for (j = 0; j < inlink->channels; j++)
 284                     s->update(s, ibuf[j]);
 285
 286                 ibuf += inlink->channels;
 287                 nb_samples_read += inlink->channels;
 288             }
 289         }
 290         break;
 291
 292     case SILENCE_TRIM_FLUSH:
 293 silence_trim_flush:
 294         nbs  = s->start_holdoff_end - s->start_holdoff_offset;
 295         nbs -= nbs % inlink->channels;
 296         if (!nbs)
 297             break;
 298
 299         out = ff_get_audio_buffer(inlink, nbs / inlink->channels);
 300         if (!out) {
 301             av_frame_free(&in);
 302             return AVERROR(ENOMEM);
 303         }
 304
 305         memcpy(out->data[0], &s->start_holdoff[s->start_holdoff_offset],
 306                nbs * sizeof(double));
 307
 308         out->pts = s->next_pts;
 309         s->next_pts += av_rescale_q(out->nb_samples,
 310                                     (AVRational){1, outlink->sample_rate},
 311                                     outlink->time_base);
 312
 313         s->start_holdoff_offset += nbs;
 314
 315         ret = ff_filter_frame(outlink, out);
 316
 317         if (s->start_holdoff_offset == s->start_holdoff_end) {
 318             s->start_holdoff_offset = 0;
 319             s->start_holdoff_end = 0;
 320             s->mode = SILENCE_COPY;
 321             goto silence_copy;
 322         }
 323         break;
 324
 325     case SILENCE_COPY:
 326 silence_copy:
 327         nbs = in->nb_samples - nb_samples_read / inlink->channels;
 328         if (!nbs)
 329             break;
 330
 331         out = ff_get_audio_buffer(inlink, nbs);
 332         if (!out) {
 333             av_frame_free(&in);
 334             return AVERROR(ENOMEM);
 335         }
 336         obuf = (double *)out->data[0];
 337
 338         if (s->stop_periods) {
 339             for (i = 0; i < nbs; i++) {
 340                 threshold = 1;
 341                 for (j = 0; j < inlink->channels; j++)
 342                     threshold &= s->compute(s, ibuf[j]) > s->stop_threshold;
 343
 344                 if (threshold && s->stop_holdoff_end && !s->leave_silence) {
 345                     s->mode = SILENCE_COPY_FLUSH;
 346                     flush(s, out, outlink, &nb_samples_written, &ret);
 347                     goto silence_copy_flush;
 348                 } else if (threshold) {
 349                     for (j = 0; j < inlink->channels; j++) {
 350                         s->update(s, *ibuf);
 351                         *obuf++ = *ibuf++;
 352                     }
 353                     nb_samples_read    += inlink->channels;
 354                     nb_samples_written += inlink->channels;
 355                 } else if (!threshold) {
 356                     for (j = 0; j < inlink->channels; j++) {
 357                         s->update(s, *ibuf);
 358                         if (s->leave_silence) {
 359                             *obuf++ = *ibuf;
 360                             nb_samples_written++;
 361                         }
 362
 363                         s->stop_holdoff[s->stop_holdoff_end++] = *ibuf++;
 364                     }
 365                     nb_samples_read += inlink->channels;
 366
 367                     if (s->stop_holdoff_end >= s->stop_duration * inlink->channels) {
 368                         if (++s->stop_found_periods >= s->stop_periods) {
 369                             s->stop_holdoff_offset = 0;
 370                             s->stop_holdoff_end = 0;
 371
 372                             if (!s->restart) {
 373                                 s->mode = SILENCE_STOP;
 374                                 flush(s, out, outlink, &nb_samples_written, &ret);
 375                                 goto silence_stop;
 376                             } else {
 377                                 s->stop_found_periods = 0;
 378                                 s->start_found_periods = 0;
 379                                 s->start_holdoff_offset = 0;
 380                                 s->start_holdoff_end = 0;
 381                                 clear_window(s);
 382                                 s->mode = SILENCE_TRIM;
 383                                 flush(s, out, outlink, &nb_samples_written, &ret);
 384                                 goto silence_trim;
 385                             }
 386                         }
 387                         s->mode = SILENCE_COPY_FLUSH;
 388                         flush(s, out, outlink, &nb_samples_written, &ret);
 389                         goto silence_copy_flush;
 390                     }
 391                 }
 392             }
 393             flush(s, out, outlink, &nb_samples_written, &ret);
 394         } else {
 395             memcpy(obuf, ibuf, sizeof(double) * nbs * inlink->channels);
 396
 397             out->pts = s->next_pts;
 398             s->next_pts += av_rescale_q(out->nb_samples,
 399                                         (AVRational){1, outlink->sample_rate},
 400                                         outlink->time_base);
 401
 402             ret = ff_filter_frame(outlink, out);
 403         }
 404         break;
 405
 406     case SILENCE_COPY_FLUSH:
 407 silence_copy_flush:
 408         nbs  = s->stop_holdoff_end - s->stop_holdoff_offset;
 409         nbs -= nbs % inlink->channels;
 410         if (!nbs)
 411             break;
 412
 413         out = ff_get_audio_buffer(inlink, nbs / inlink->channels);
 414         if (!out) {
 415             av_frame_free(&in);
 416             return AVERROR(ENOMEM);
 417         }
 418
 419         memcpy(out->data[0], &s->stop_holdoff[s->stop_holdoff_offset],
 420                nbs * sizeof(double));
 421         s->stop_holdoff_offset += nbs;
 422
 423         out->pts = s->next_pts;
 424         s->next_pts += av_rescale_q(out->nb_samples,
 425                                     (AVRational){1, outlink->sample_rate},
 426                                     outlink->time_base);
 427
 428         ret = ff_filter_frame(outlink, out);
 429
 430         if (s->stop_holdoff_offset == s->stop_holdoff_end) {
 431             s->stop_holdoff_offset = 0;
 432             s->stop_holdoff_end = 0;
 433             s->mode = SILENCE_COPY;
 434             goto silence_copy;
 435         }
 436         break;
 437     case SILENCE_STOP:
 438 silence_stop:
 439         break;
 440     }
 441
 442     av_frame_free(&in);
 443
 444     return ret;
 445 }
 446
 447 static int request_frame(AVFilterLink *outlink)
 448 {
 449     AVFilterContext *ctx = outlink->src;
 450     SilenceRemoveContext *s = ctx->priv;
 451     int ret;
 452
 453     ret = ff_request_frame(ctx->inputs[0]);
 454     if (ret == AVERROR_EOF && (s->mode == SILENCE_COPY_FLUSH ||
 455                                s->mode == SILENCE_COPY)) {
 456         int nbs = s->stop_holdoff_end - s->stop_holdoff_offset;
 457         if (nbs) {
 458             AVFrame *frame;
 459
 460             frame = ff_get_audio_buffer(outlink, nbs / outlink->channels);
 461             if (!frame)
 462                 return AVERROR(ENOMEM);
 463
 464             memcpy(frame->data[0], &s->stop_holdoff[s->stop_holdoff_offset],
 465                    nbs * sizeof(double));
 466
 467             frame->pts = s->next_pts;
 468             s->next_pts += av_rescale_q(frame->nb_samples,
 469                                         (AVRational){1, outlink->sample_rate},
 470                                         outlink->time_base);
 471
 472             ret = ff_filter_frame(ctx->inputs[0], frame);
 473         }
 474         s->mode = SILENCE_STOP;
 475     }
 476     return ret;
 477 }
 478
 479 static int query_formats(AVFilterContext *ctx)
 480 {
 481     AVFilterFormats *formats = NULL;
 482     AVFilterChannelLayouts *layouts = NULL;
 483     static const enum AVSampleFormat sample_fmts[] = {
 484         AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_NONE
 485     };
 486     int ret;
 487
 488     layouts = ff_all_channel_counts();
 489     if (!layouts)
 490         return AVERROR(ENOMEM);
 491     ret = ff_set_common_channel_layouts(ctx, layouts);
 492     if (ret < 0)
 493         return ret;
 494
 495     formats = ff_make_format_list(sample_fmts);
 496     if (!formats)
 497         return AVERROR(ENOMEM);
 498     ret = ff_set_common_formats(ctx, formats);
 499     if (ret < 0)
 500         return ret;
 501
 502     formats = ff_all_samplerates();
 503     if (!formats)
 504         return AVERROR(ENOMEM);
 505     return ff_set_common_samplerates(ctx, formats);
 506 }
 507
 508 static av_cold void uninit(AVFilterContext *ctx)
 509 {
 510     SilenceRemoveContext *s = ctx->priv;
 511
 512     av_freep(&s->start_holdoff);
 513     av_freep(&s->stop_holdoff);
 514     av_freep(&s->window);
 515 }
 516
 517 static const AVFilterPad silenceremove_inputs[] = {
 518     {
 519         .name         = "default",
 520         .type         = AVMEDIA_TYPE_AUDIO,
 521         .config_props = config_input,
 522         .filter_frame = filter_frame,
 523     },
 524     { NULL }
 525 };
 526
 527 static const AVFilterPad silenceremove_outputs[] = {
 528     {
 529         .name          = "default",
 530         .type          = AVMEDIA_TYPE_AUDIO,
 531         .request_frame = request_frame,
 532     },
 533     { NULL }
 534 };
 535
 536 AVFilter ff_af_silenceremove = {
 537     .name          = "silenceremove",
 538     .description   = NULL_IF_CONFIG_SMALL("Remove silence."),
 539     .priv_size     = sizeof(SilenceRemoveContext),
 540     .priv_class    = &silenceremove_class,
 541     .init          = init,
 542     .uninit        = uninit,
 543     .query_formats = query_formats,
 544     .inputs        = silenceremove_inputs,
 545     .outputs       = silenceremove_outputs,
 546 };