| 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; |
| 65 | double *window_current; |
| 66 | double *window_end; |
| 67 | int window_size; |
| 68 | double rms_sum; |
| 69 | |
| 70 | int leave_silence; |
| 71 | int restart; |
| 72 | int64_t next_pts; |
| 73 | } SilenceRemoveContext; |
| 74 | |
| 75 | #define OFFSET(x) offsetof(SilenceRemoveContext, x) |
| 76 | #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_AUDIO_PARAM |
| 77 | static const AVOption silenceremove_options[] = { |
| 78 | { "start_periods", NULL, OFFSET(start_periods), AV_OPT_TYPE_INT, {.i64=0}, 0, 9000, FLAGS }, |
| 79 | { "start_duration", NULL, OFFSET(start_duration), AV_OPT_TYPE_DURATION, {.i64=0}, 0, 9000, FLAGS }, |
| 80 | { "start_threshold", NULL, OFFSET(start_threshold), AV_OPT_TYPE_DOUBLE, {.dbl=0}, 0, DBL_MAX, FLAGS }, |
| 81 | { "stop_periods", NULL, OFFSET(stop_periods), AV_OPT_TYPE_INT, {.i64=0}, -9000, 9000, FLAGS }, |
| 82 | { "stop_duration", NULL, OFFSET(stop_duration), AV_OPT_TYPE_DURATION, {.i64=0}, 0, 9000, FLAGS }, |
| 83 | { "stop_threshold", NULL, OFFSET(stop_threshold), AV_OPT_TYPE_DOUBLE, {.dbl=0}, 0, DBL_MAX, FLAGS }, |
| 84 | { "leave_silence", NULL, OFFSET(leave_silence), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, FLAGS }, |
| 85 | { NULL } |
| 86 | }; |
| 87 | |
| 88 | AVFILTER_DEFINE_CLASS(silenceremove); |
| 89 | |
| 90 | static av_cold int init(AVFilterContext *ctx) |
| 91 | { |
| 92 | SilenceRemoveContext *s = ctx->priv; |
| 93 | |
| 94 | if (s->stop_periods < 0) { |
| 95 | s->stop_periods = -s->stop_periods; |
| 96 | s->restart = 1; |
| 97 | } |
| 98 | |
| 99 | return 0; |
| 100 | } |
| 101 | |
| 102 | static void clear_rms(SilenceRemoveContext *s) |
| 103 | { |
| 104 | memset(s->window, 0, s->window_size * sizeof(*s->window)); |
| 105 | |
| 106 | s->window_current = s->window; |
| 107 | s->window_end = s->window + s->window_size; |
| 108 | s->rms_sum = 0; |
| 109 | } |
| 110 | |
| 111 | static int config_input(AVFilterLink *inlink) |
| 112 | { |
| 113 | AVFilterContext *ctx = inlink->dst; |
| 114 | SilenceRemoveContext *s = ctx->priv; |
| 115 | |
| 116 | s->window_size = (inlink->sample_rate / 50) * inlink->channels; |
| 117 | s->window = av_malloc_array(s->window_size, sizeof(*s->window)); |
| 118 | if (!s->window) |
| 119 | return AVERROR(ENOMEM); |
| 120 | |
| 121 | clear_rms(s); |
| 122 | |
| 123 | s->start_duration = av_rescale(s->start_duration, inlink->sample_rate, |
| 124 | AV_TIME_BASE); |
| 125 | s->stop_duration = av_rescale(s->stop_duration, inlink->sample_rate, |
| 126 | AV_TIME_BASE); |
| 127 | |
| 128 | s->start_holdoff = av_malloc_array(FFMAX(s->start_duration, 1), |
| 129 | sizeof(*s->start_holdoff) * |
| 130 | inlink->channels); |
| 131 | if (!s->start_holdoff) |
| 132 | return AVERROR(ENOMEM); |
| 133 | |
| 134 | s->start_holdoff_offset = 0; |
| 135 | s->start_holdoff_end = 0; |
| 136 | s->start_found_periods = 0; |
| 137 | |
| 138 | s->stop_holdoff = av_malloc_array(FFMAX(s->stop_duration, 1), |
| 139 | sizeof(*s->stop_holdoff) * |
| 140 | inlink->channels); |
| 141 | if (!s->stop_holdoff) |
| 142 | return AVERROR(ENOMEM); |
| 143 | |
| 144 | s->stop_holdoff_offset = 0; |
| 145 | s->stop_holdoff_end = 0; |
| 146 | s->stop_found_periods = 0; |
| 147 | |
| 148 | if (s->start_periods) |
| 149 | s->mode = SILENCE_TRIM; |
| 150 | else |
| 151 | s->mode = SILENCE_COPY; |
| 152 | |
| 153 | return 0; |
| 154 | } |
| 155 | |
| 156 | static int config_output(AVFilterLink *outlink) |
| 157 | { |
| 158 | outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP; |
| 159 | |
| 160 | return 0; |
| 161 | } |
| 162 | |
| 163 | static double compute_rms(SilenceRemoveContext *s, double sample) |
| 164 | { |
| 165 | double new_sum; |
| 166 | |
| 167 | new_sum = s->rms_sum; |
| 168 | new_sum -= *s->window_current; |
| 169 | new_sum += sample * sample; |
| 170 | |
| 171 | return sqrt(new_sum / s->window_size); |
| 172 | } |
| 173 | |
| 174 | static void update_rms(SilenceRemoveContext *s, double sample) |
| 175 | { |
| 176 | s->rms_sum -= *s->window_current; |
| 177 | *s->window_current = sample * sample; |
| 178 | s->rms_sum += *s->window_current; |
| 179 | |
| 180 | s->window_current++; |
| 181 | if (s->window_current >= s->window_end) |
| 182 | s->window_current = s->window; |
| 183 | } |
| 184 | |
| 185 | static void flush(AVFrame *out, AVFilterLink *outlink, |
| 186 | int *nb_samples_written, int *ret) |
| 187 | { |
| 188 | if (*nb_samples_written) { |
| 189 | out->nb_samples = *nb_samples_written / outlink->channels; |
| 190 | *ret = ff_filter_frame(outlink, out); |
| 191 | *nb_samples_written = 0; |
| 192 | } else { |
| 193 | av_frame_free(&out); |
| 194 | } |
| 195 | } |
| 196 | |
| 197 | static int filter_frame(AVFilterLink *inlink, AVFrame *in) |
| 198 | { |
| 199 | AVFilterContext *ctx = inlink->dst; |
| 200 | AVFilterLink *outlink = ctx->outputs[0]; |
| 201 | SilenceRemoveContext *s = ctx->priv; |
| 202 | int i, j, threshold, ret = 0; |
| 203 | int nbs, nb_samples_read, nb_samples_written; |
| 204 | double *obuf, *ibuf = (double *)in->data[0]; |
| 205 | AVFrame *out; |
| 206 | |
| 207 | nb_samples_read = nb_samples_written = 0; |
| 208 | |
| 209 | switch (s->mode) { |
| 210 | case SILENCE_TRIM: |
| 211 | silence_trim: |
| 212 | nbs = in->nb_samples - nb_samples_read / inlink->channels; |
| 213 | if (!nbs) |
| 214 | break; |
| 215 | |
| 216 | for (i = 0; i < nbs; i++) { |
| 217 | threshold = 0; |
| 218 | for (j = 0; j < inlink->channels; j++) { |
| 219 | threshold |= compute_rms(s, ibuf[j]) > s->start_threshold; |
| 220 | } |
| 221 | |
| 222 | if (threshold) { |
| 223 | for (j = 0; j < inlink->channels; j++) { |
| 224 | update_rms(s, *ibuf); |
| 225 | s->start_holdoff[s->start_holdoff_end++] = *ibuf++; |
| 226 | nb_samples_read++; |
| 227 | } |
| 228 | |
| 229 | if (s->start_holdoff_end >= s->start_duration * inlink->channels) { |
| 230 | if (++s->start_found_periods >= s->start_periods) { |
| 231 | s->mode = SILENCE_TRIM_FLUSH; |
| 232 | goto silence_trim_flush; |
| 233 | } |
| 234 | |
| 235 | s->start_holdoff_offset = 0; |
| 236 | s->start_holdoff_end = 0; |
| 237 | } |
| 238 | } else { |
| 239 | s->start_holdoff_end = 0; |
| 240 | |
| 241 | for (j = 0; j < inlink->channels; j++) |
| 242 | update_rms(s, ibuf[j]); |
| 243 | |
| 244 | ibuf += inlink->channels; |
| 245 | nb_samples_read += inlink->channels; |
| 246 | } |
| 247 | } |
| 248 | break; |
| 249 | |
| 250 | case SILENCE_TRIM_FLUSH: |
| 251 | silence_trim_flush: |
| 252 | nbs = s->start_holdoff_end - s->start_holdoff_offset; |
| 253 | nbs -= nbs % inlink->channels; |
| 254 | if (!nbs) |
| 255 | break; |
| 256 | |
| 257 | out = ff_get_audio_buffer(inlink, nbs / inlink->channels); |
| 258 | if (!out) { |
| 259 | av_frame_free(&in); |
| 260 | return AVERROR(ENOMEM); |
| 261 | } |
| 262 | |
| 263 | memcpy(out->data[0], &s->start_holdoff[s->start_holdoff_offset], |
| 264 | nbs * sizeof(double)); |
| 265 | s->start_holdoff_offset += nbs; |
| 266 | |
| 267 | ret = ff_filter_frame(outlink, out); |
| 268 | |
| 269 | if (s->start_holdoff_offset == s->start_holdoff_end) { |
| 270 | s->start_holdoff_offset = 0; |
| 271 | s->start_holdoff_end = 0; |
| 272 | s->mode = SILENCE_COPY; |
| 273 | goto silence_copy; |
| 274 | } |
| 275 | break; |
| 276 | |
| 277 | case SILENCE_COPY: |
| 278 | silence_copy: |
| 279 | nbs = in->nb_samples - nb_samples_read / inlink->channels; |
| 280 | if (!nbs) |
| 281 | break; |
| 282 | |
| 283 | out = ff_get_audio_buffer(inlink, nbs); |
| 284 | if (!out) { |
| 285 | av_frame_free(&in); |
| 286 | return AVERROR(ENOMEM); |
| 287 | } |
| 288 | obuf = (double *)out->data[0]; |
| 289 | |
| 290 | if (s->stop_periods) { |
| 291 | for (i = 0; i < nbs; i++) { |
| 292 | threshold = 1; |
| 293 | for (j = 0; j < inlink->channels; j++) |
| 294 | threshold &= compute_rms(s, ibuf[j]) > s->stop_threshold; |
| 295 | |
| 296 | if (threshold && s->stop_holdoff_end && !s->leave_silence) { |
| 297 | s->mode = SILENCE_COPY_FLUSH; |
| 298 | flush(out, outlink, &nb_samples_written, &ret); |
| 299 | goto silence_copy_flush; |
| 300 | } else if (threshold) { |
| 301 | for (j = 0; j < inlink->channels; j++) { |
| 302 | update_rms(s, *ibuf); |
| 303 | *obuf++ = *ibuf++; |
| 304 | nb_samples_read++; |
| 305 | nb_samples_written++; |
| 306 | } |
| 307 | } else if (!threshold) { |
| 308 | for (j = 0; j < inlink->channels; j++) { |
| 309 | update_rms(s, *ibuf); |
| 310 | if (s->leave_silence) { |
| 311 | *obuf++ = *ibuf; |
| 312 | nb_samples_written++; |
| 313 | } |
| 314 | |
| 315 | s->stop_holdoff[s->stop_holdoff_end++] = *ibuf++; |
| 316 | nb_samples_read++; |
| 317 | } |
| 318 | |
| 319 | if (s->stop_holdoff_end >= s->stop_duration * inlink->channels) { |
| 320 | if (++s->stop_found_periods >= s->stop_periods) { |
| 321 | s->stop_holdoff_offset = 0; |
| 322 | s->stop_holdoff_end = 0; |
| 323 | |
| 324 | if (!s->restart) { |
| 325 | s->mode = SILENCE_STOP; |
| 326 | flush(out, outlink, &nb_samples_written, &ret); |
| 327 | goto silence_stop; |
| 328 | } else { |
| 329 | s->stop_found_periods = 0; |
| 330 | s->start_found_periods = 0; |
| 331 | s->start_holdoff_offset = 0; |
| 332 | s->start_holdoff_end = 0; |
| 333 | clear_rms(s); |
| 334 | s->mode = SILENCE_TRIM; |
| 335 | flush(out, outlink, &nb_samples_written, &ret); |
| 336 | goto silence_trim; |
| 337 | } |
| 338 | } |
| 339 | s->mode = SILENCE_COPY_FLUSH; |
| 340 | flush(out, outlink, &nb_samples_written, &ret); |
| 341 | goto silence_copy_flush; |
| 342 | } |
| 343 | } |
| 344 | } |
| 345 | flush(out, outlink, &nb_samples_written, &ret); |
| 346 | } else { |
| 347 | memcpy(obuf, ibuf, sizeof(double) * nbs * inlink->channels); |
| 348 | ret = ff_filter_frame(outlink, out); |
| 349 | } |
| 350 | break; |
| 351 | |
| 352 | case SILENCE_COPY_FLUSH: |
| 353 | silence_copy_flush: |
| 354 | nbs = s->stop_holdoff_end - s->stop_holdoff_offset; |
| 355 | nbs -= nbs % inlink->channels; |
| 356 | if (!nbs) |
| 357 | break; |
| 358 | |
| 359 | out = ff_get_audio_buffer(inlink, nbs / inlink->channels); |
| 360 | if (!out) { |
| 361 | av_frame_free(&in); |
| 362 | return AVERROR(ENOMEM); |
| 363 | } |
| 364 | |
| 365 | memcpy(out->data[0], &s->stop_holdoff[s->stop_holdoff_offset], |
| 366 | nbs * sizeof(double)); |
| 367 | s->stop_holdoff_offset += nbs; |
| 368 | |
| 369 | ret = ff_filter_frame(outlink, out); |
| 370 | |
| 371 | if (s->stop_holdoff_offset == s->stop_holdoff_end) { |
| 372 | s->stop_holdoff_offset = 0; |
| 373 | s->stop_holdoff_end = 0; |
| 374 | s->mode = SILENCE_COPY; |
| 375 | goto silence_copy; |
| 376 | } |
| 377 | break; |
| 378 | case SILENCE_STOP: |
| 379 | silence_stop: |
| 380 | break; |
| 381 | } |
| 382 | |
| 383 | av_frame_free(&in); |
| 384 | |
| 385 | return ret; |
| 386 | } |
| 387 | |
| 388 | static int request_frame(AVFilterLink *outlink) |
| 389 | { |
| 390 | AVFilterContext *ctx = outlink->src; |
| 391 | SilenceRemoveContext *s = ctx->priv; |
| 392 | int ret; |
| 393 | |
| 394 | ret = ff_request_frame(ctx->inputs[0]); |
| 395 | if (ret == AVERROR_EOF && (s->mode == SILENCE_COPY_FLUSH || |
| 396 | s->mode == SILENCE_COPY)) { |
| 397 | int nbs = s->stop_holdoff_end - s->stop_holdoff_offset; |
| 398 | if (nbs) { |
| 399 | AVFrame *frame; |
| 400 | |
| 401 | frame = ff_get_audio_buffer(outlink, nbs / outlink->channels); |
| 402 | if (!frame) |
| 403 | return AVERROR(ENOMEM); |
| 404 | |
| 405 | memcpy(frame->data[0], &s->stop_holdoff[s->stop_holdoff_offset], |
| 406 | nbs * sizeof(double)); |
| 407 | ret = ff_filter_frame(ctx->inputs[0], frame); |
| 408 | } |
| 409 | s->mode = SILENCE_STOP; |
| 410 | } |
| 411 | return ret; |
| 412 | } |
| 413 | |
| 414 | static int query_formats(AVFilterContext *ctx) |
| 415 | { |
| 416 | AVFilterFormats *formats = NULL; |
| 417 | AVFilterChannelLayouts *layouts = NULL; |
| 418 | static const enum AVSampleFormat sample_fmts[] = { |
| 419 | AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_NONE |
| 420 | }; |
| 421 | |
| 422 | layouts = ff_all_channel_layouts(); |
| 423 | if (!layouts) |
| 424 | return AVERROR(ENOMEM); |
| 425 | ff_set_common_channel_layouts(ctx, layouts); |
| 426 | |
| 427 | formats = ff_make_format_list(sample_fmts); |
| 428 | if (!formats) |
| 429 | return AVERROR(ENOMEM); |
| 430 | ff_set_common_formats(ctx, formats); |
| 431 | |
| 432 | formats = ff_all_samplerates(); |
| 433 | if (!formats) |
| 434 | return AVERROR(ENOMEM); |
| 435 | ff_set_common_samplerates(ctx, formats); |
| 436 | |
| 437 | return 0; |
| 438 | } |
| 439 | |
| 440 | static av_cold void uninit(AVFilterContext *ctx) |
| 441 | { |
| 442 | SilenceRemoveContext *s = ctx->priv; |
| 443 | |
| 444 | av_freep(&s->start_holdoff); |
| 445 | av_freep(&s->stop_holdoff); |
| 446 | av_freep(&s->window); |
| 447 | } |
| 448 | |
| 449 | static const AVFilterPad silenceremove_inputs[] = { |
| 450 | { |
| 451 | .name = "default", |
| 452 | .type = AVMEDIA_TYPE_AUDIO, |
| 453 | .config_props = config_input, |
| 454 | .filter_frame = filter_frame, |
| 455 | }, |
| 456 | { NULL } |
| 457 | }; |
| 458 | |
| 459 | static const AVFilterPad silenceremove_outputs[] = { |
| 460 | { |
| 461 | .name = "default", |
| 462 | .type = AVMEDIA_TYPE_AUDIO, |
| 463 | .config_props = config_output, |
| 464 | .request_frame = request_frame, |
| 465 | }, |
| 466 | { NULL } |
| 467 | }; |
| 468 | |
| 469 | AVFilter ff_af_silenceremove = { |
| 470 | .name = "silenceremove", |
| 471 | .description = NULL_IF_CONFIG_SMALL("Remove silence."), |
| 472 | .priv_size = sizeof(SilenceRemoveContext), |
| 473 | .priv_class = &silenceremove_class, |
| 474 | .init = init, |
| 475 | .uninit = uninit, |
| 476 | .query_formats = query_formats, |
| 477 | .inputs = silenceremove_inputs, |
| 478 | .outputs = silenceremove_outputs, |
| 479 | }; |