FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
af_compand.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 1999 Chris Bagwell
3  * Copyright (c) 1999 Nick Bailey
4  * Copyright (c) 2007 Rob Sykes <robs@users.sourceforge.net>
5  * Copyright (c) 2013 Paul B Mahol
6  * Copyright (c) 2014 Andrew Kelley
7  *
8  * This file is part of FFmpeg.
9  *
10  * FFmpeg is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU Lesser General Public
12  * License as published by the Free Software Foundation; either
13  * version 2.1 of the License, or (at your option) any later version.
14  *
15  * FFmpeg is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18  * Lesser General Public License for more details.
19  *
20  * You should have received a copy of the GNU Lesser General Public
21  * License along with FFmpeg; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23  */
24 
25 /**
26  * @file
27  * audio compand filter
28  */
29 
30 #include "libavutil/avassert.h"
31 #include "libavutil/avstring.h"
32 #include "libavutil/opt.h"
33 #include "libavutil/samplefmt.h"
34 #include "audio.h"
35 #include "avfilter.h"
36 #include "internal.h"
37 
38 typedef struct ChanParam {
39  double attack;
40  double decay;
41  double volume;
42 } ChanParam;
43 
44 typedef struct CompandSegment {
45  double x, y;
46  double a, b;
48 
49 typedef struct CompandContext {
50  const AVClass *class;
52  char *attacks, *decays, *points;
55  double in_min_lin;
56  double out_min_lin;
57  double curve_dB;
58  double gain_dB;
60  double delay;
65  int64_t pts;
66 
69 
70 #define OFFSET(x) offsetof(CompandContext, x)
71 #define A AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
72 
73 static const AVOption compand_options[] = {
74  { "attacks", "set time over which increase of volume is determined", OFFSET(attacks), AV_OPT_TYPE_STRING, { .str = "0.3" }, 0, 0, A },
75  { "decays", "set time over which decrease of volume is determined", OFFSET(decays), AV_OPT_TYPE_STRING, { .str = "0.8" }, 0, 0, A },
76  { "points", "set points of transfer function", OFFSET(points), AV_OPT_TYPE_STRING, { .str = "-70/-70|-60/-20" }, 0, 0, A },
77  { "soft-knee", "set soft-knee", OFFSET(curve_dB), AV_OPT_TYPE_DOUBLE, { .dbl = 0.01 }, 0.01, 900, A },
78  { "gain", "set output gain", OFFSET(gain_dB), AV_OPT_TYPE_DOUBLE, { .dbl = 0 }, -900, 900, A },
79  { "volume", "set initial volume", OFFSET(initial_volume), AV_OPT_TYPE_DOUBLE, { .dbl = 0 }, -900, 0, A },
80  { "delay", "set delay for samples before sending them to volume adjuster", OFFSET(delay), AV_OPT_TYPE_DOUBLE, { .dbl = 0 }, 0, 20, A },
81  { NULL }
82 };
83 
84 AVFILTER_DEFINE_CLASS(compand);
85 
86 static av_cold int init(AVFilterContext *ctx)
87 {
88  CompandContext *s = ctx->priv;
89  s->pts = AV_NOPTS_VALUE;
90  return 0;
91 }
92 
93 static av_cold void uninit(AVFilterContext *ctx)
94 {
95  CompandContext *s = ctx->priv;
96 
97  av_freep(&s->channels);
98  av_freep(&s->segments);
100 }
101 
103 {
106  static const enum AVSampleFormat sample_fmts[] = {
109  };
110 
111  layouts = ff_all_channel_layouts();
112  if (!layouts)
113  return AVERROR(ENOMEM);
114  ff_set_common_channel_layouts(ctx, layouts);
115 
116  formats = ff_make_format_list(sample_fmts);
117  if (!formats)
118  return AVERROR(ENOMEM);
119  ff_set_common_formats(ctx, formats);
120 
121  formats = ff_all_samplerates();
122  if (!formats)
123  return AVERROR(ENOMEM);
124  ff_set_common_samplerates(ctx, formats);
125 
126  return 0;
127 }
128 
129 static void count_items(char *item_str, int *nb_items)
130 {
131  char *p;
132 
133  *nb_items = 1;
134  for (p = item_str; *p; p++) {
135  if (*p == ' ' || *p == '|')
136  (*nb_items)++;
137  }
138 }
139 
140 static void update_volume(ChanParam *cp, double in)
141 {
142  double delta = in - cp->volume;
143 
144  if (delta > 0.0)
145  cp->volume += delta * cp->attack;
146  else
147  cp->volume += delta * cp->decay;
148 }
149 
150 static double get_volume(CompandContext *s, double in_lin)
151 {
152  CompandSegment *cs;
153  double in_log, out_log;
154  int i;
155 
156  if (in_lin < s->in_min_lin)
157  return s->out_min_lin;
158 
159  in_log = log(in_lin);
160 
161  for (i = 1; i < s->nb_segments; i++)
162  if (in_log <= s->segments[i].x)
163  break;
164  cs = &s->segments[i - 1];
165  in_log -= cs->x;
166  out_log = cs->y + in_log * (cs->a * in_log + cs->b);
167 
168  return exp(out_log);
169 }
170 
172 {
173  CompandContext *s = ctx->priv;
174  AVFilterLink *inlink = ctx->inputs[0];
175  const int channels = inlink->channels;
176  const int nb_samples = frame->nb_samples;
177  AVFrame *out_frame;
178  int chan, i;
179  int err;
180 
181  if (av_frame_is_writable(frame)) {
182  out_frame = frame;
183  } else {
184  out_frame = ff_get_audio_buffer(inlink, nb_samples);
185  if (!out_frame) {
186  av_frame_free(&frame);
187  return AVERROR(ENOMEM);
188  }
189  err = av_frame_copy_props(out_frame, frame);
190  if (err < 0) {
191  av_frame_free(&out_frame);
192  av_frame_free(&frame);
193  return err;
194  }
195  }
196 
197  for (chan = 0; chan < channels; chan++) {
198  const double *src = (double *)frame->extended_data[chan];
199  double *dst = (double *)out_frame->extended_data[chan];
200  ChanParam *cp = &s->channels[chan];
201 
202  for (i = 0; i < nb_samples; i++) {
203  update_volume(cp, fabs(src[i]));
204 
205  dst[i] = av_clipd(src[i] * get_volume(s, cp->volume), -1, 1);
206  }
207  }
208 
209  if (frame != out_frame)
210  av_frame_free(&frame);
211 
212  return ff_filter_frame(ctx->outputs[0], out_frame);
213 }
214 
215 #define MOD(a, b) (((a) >= (b)) ? (a) - (b) : (a))
216 
218 {
219  CompandContext *s = ctx->priv;
220  AVFilterLink *inlink = ctx->inputs[0];
221  const int channels = inlink->channels;
222  const int nb_samples = frame->nb_samples;
223  int chan, i, av_uninit(dindex), oindex, av_uninit(count);
224  AVFrame *out_frame = NULL;
225  int err;
226 
227  if (s->pts == AV_NOPTS_VALUE) {
228  s->pts = (frame->pts == AV_NOPTS_VALUE) ? 0 : frame->pts;
229  }
230 
231  av_assert1(channels > 0); /* would corrupt delay_count and delay_index */
232 
233  for (chan = 0; chan < channels; chan++) {
234  AVFrame *delay_frame = s->delay_frame;
235  const double *src = (double *)frame->extended_data[chan];
236  double *dbuf = (double *)delay_frame->extended_data[chan];
237  ChanParam *cp = &s->channels[chan];
238  double *dst;
239 
240  count = s->delay_count;
241  dindex = s->delay_index;
242  for (i = 0, oindex = 0; i < nb_samples; i++) {
243  const double in = src[i];
244  update_volume(cp, fabs(in));
245 
246  if (count >= s->delay_samples) {
247  if (!out_frame) {
248  out_frame = ff_get_audio_buffer(inlink, nb_samples - i);
249  if (!out_frame) {
250  av_frame_free(&frame);
251  return AVERROR(ENOMEM);
252  }
253  err = av_frame_copy_props(out_frame, frame);
254  if (err < 0) {
255  av_frame_free(&out_frame);
256  av_frame_free(&frame);
257  return err;
258  }
259  out_frame->pts = s->pts;
260  s->pts += av_rescale_q(nb_samples - i,
261  (AVRational){ 1, inlink->sample_rate },
262  inlink->time_base);
263  }
264 
265  dst = (double *)out_frame->extended_data[chan];
266  dst[oindex++] = av_clipd(dbuf[dindex] *
267  get_volume(s, cp->volume), -1, 1);
268  } else {
269  count++;
270  }
271 
272  dbuf[dindex] = in;
273  dindex = MOD(dindex + 1, s->delay_samples);
274  }
275  }
276 
277  s->delay_count = count;
278  s->delay_index = dindex;
279 
280  av_frame_free(&frame);
281 
282  if (out_frame) {
283  err = ff_filter_frame(ctx->outputs[0], out_frame);
284  return err;
285  }
286 
287  return 0;
288 }
289 
290 static int compand_drain(AVFilterLink *outlink)
291 {
292  AVFilterContext *ctx = outlink->src;
293  CompandContext *s = ctx->priv;
294  const int channels = outlink->channels;
295  AVFrame *frame = NULL;
296  int chan, i, dindex;
297 
298  /* 2048 is to limit output frame size during drain */
299  frame = ff_get_audio_buffer(outlink, FFMIN(2048, s->delay_count));
300  if (!frame)
301  return AVERROR(ENOMEM);
302  frame->pts = s->pts;
303  s->pts += av_rescale_q(frame->nb_samples,
304  (AVRational){ 1, outlink->sample_rate }, outlink->time_base);
305 
306  av_assert0(channels > 0);
307  for (chan = 0; chan < channels; chan++) {
308  AVFrame *delay_frame = s->delay_frame;
309  double *dbuf = (double *)delay_frame->extended_data[chan];
310  double *dst = (double *)frame->extended_data[chan];
311  ChanParam *cp = &s->channels[chan];
312 
313  dindex = s->delay_index;
314  for (i = 0; i < frame->nb_samples; i++) {
315  dst[i] = av_clipd(dbuf[dindex] * get_volume(s, cp->volume),
316  -1, 1);
317  dindex = MOD(dindex + 1, s->delay_samples);
318  }
319  }
320  s->delay_count -= frame->nb_samples;
321  s->delay_index = dindex;
322 
323  return ff_filter_frame(outlink, frame);
324 }
325 
326 static int config_output(AVFilterLink *outlink)
327 {
328  AVFilterContext *ctx = outlink->src;
329  CompandContext *s = ctx->priv;
330  const int sample_rate = outlink->sample_rate;
331  double radius = s->curve_dB * M_LN10 / 20.0;
332  char *p, *saveptr = NULL;
333  const int channels = outlink->channels;
334  int nb_attacks, nb_decays, nb_points;
335  int new_nb_items, num;
336  int i;
337  int err;
338 
339 
340  count_items(s->attacks, &nb_attacks);
341  count_items(s->decays, &nb_decays);
342  count_items(s->points, &nb_points);
343 
344  if (channels <= 0) {
345  av_log(ctx, AV_LOG_ERROR, "Invalid number of channels: %d\n", channels);
346  return AVERROR(EINVAL);
347  }
348 
349  if (nb_attacks > channels || nb_decays > channels) {
350  av_log(ctx, AV_LOG_ERROR,
351  "Number of attacks/decays bigger than number of channels.\n");
352  return AVERROR(EINVAL);
353  }
354 
355  uninit(ctx);
356 
357  s->channels = av_mallocz_array(channels, sizeof(*s->channels));
358  s->nb_segments = (nb_points + 4) * 2;
359  s->segments = av_mallocz_array(s->nb_segments, sizeof(*s->segments));
360 
361  if (!s->channels || !s->segments) {
362  uninit(ctx);
363  return AVERROR(ENOMEM);
364  }
365 
366  p = s->attacks;
367  for (i = 0, new_nb_items = 0; i < nb_attacks; i++) {
368  char *tstr = av_strtok(p, " |", &saveptr);
369  p = NULL;
370  new_nb_items += sscanf(tstr, "%lf", &s->channels[i].attack) == 1;
371  if (s->channels[i].attack < 0) {
372  uninit(ctx);
373  return AVERROR(EINVAL);
374  }
375  }
376  nb_attacks = new_nb_items;
377 
378  p = s->decays;
379  for (i = 0, new_nb_items = 0; i < nb_decays; i++) {
380  char *tstr = av_strtok(p, " |", &saveptr);
381  p = NULL;
382  new_nb_items += sscanf(tstr, "%lf", &s->channels[i].decay) == 1;
383  if (s->channels[i].decay < 0) {
384  uninit(ctx);
385  return AVERROR(EINVAL);
386  }
387  }
388  nb_decays = new_nb_items;
389 
390  if (nb_attacks != nb_decays) {
391  av_log(ctx, AV_LOG_ERROR,
392  "Number of attacks %d differs from number of decays %d.\n",
393  nb_attacks, nb_decays);
394  uninit(ctx);
395  return AVERROR(EINVAL);
396  }
397 
398 #define S(x) s->segments[2 * ((x) + 1)]
399  p = s->points;
400  for (i = 0, new_nb_items = 0; i < nb_points; i++) {
401  char *tstr = av_strtok(p, " |", &saveptr);
402  p = NULL;
403  if (sscanf(tstr, "%lf/%lf", &S(i).x, &S(i).y) != 2) {
404  av_log(ctx, AV_LOG_ERROR,
405  "Invalid and/or missing input/output value.\n");
406  uninit(ctx);
407  return AVERROR(EINVAL);
408  }
409  if (i && S(i - 1).x > S(i).x) {
410  av_log(ctx, AV_LOG_ERROR,
411  "Transfer function input values must be increasing.\n");
412  uninit(ctx);
413  return AVERROR(EINVAL);
414  }
415  S(i).y -= S(i).x;
416  av_log(ctx, AV_LOG_DEBUG, "%d: x=%f y=%f\n", i, S(i).x, S(i).y);
417  new_nb_items++;
418  }
419  num = new_nb_items;
420 
421  /* Add 0,0 if necessary */
422  if (num == 0 || S(num - 1).x)
423  num++;
424 
425 #undef S
426 #define S(x) s->segments[2 * (x)]
427  /* Add a tail off segment at the start */
428  S(0).x = S(1).x - 2 * s->curve_dB;
429  S(0).y = S(1).y;
430  num++;
431 
432  /* Join adjacent colinear segments */
433  for (i = 2; i < num; i++) {
434  double g1 = (S(i - 1).y - S(i - 2).y) * (S(i - 0).x - S(i - 1).x);
435  double g2 = (S(i - 0).y - S(i - 1).y) * (S(i - 1).x - S(i - 2).x);
436  int j;
437 
438  if (fabs(g1 - g2))
439  continue;
440  num--;
441  for (j = --i; j < num; j++)
442  S(j) = S(j + 1);
443  }
444 
445  for (i = 0; !i || s->segments[i - 2].x; i += 2) {
446  s->segments[i].y += s->gain_dB;
447  s->segments[i].x *= M_LN10 / 20;
448  s->segments[i].y *= M_LN10 / 20;
449  }
450 
451 #define L(x) s->segments[i - (x)]
452  for (i = 4; s->segments[i - 2].x; i += 2) {
453  double x, y, cx, cy, in1, in2, out1, out2, theta, len, r;
454 
455  L(4).a = 0;
456  L(4).b = (L(2).y - L(4).y) / (L(2).x - L(4).x);
457 
458  L(2).a = 0;
459  L(2).b = (L(0).y - L(2).y) / (L(0).x - L(2).x);
460 
461  theta = atan2(L(2).y - L(4).y, L(2).x - L(4).x);
462  len = sqrt(pow(L(2).x - L(4).x, 2.) + pow(L(2).y - L(4).y, 2.));
463  r = FFMIN(radius, len);
464  L(3).x = L(2).x - r * cos(theta);
465  L(3).y = L(2).y - r * sin(theta);
466 
467  theta = atan2(L(0).y - L(2).y, L(0).x - L(2).x);
468  len = sqrt(pow(L(0).x - L(2).x, 2.) + pow(L(0).y - L(2).y, 2.));
469  r = FFMIN(radius, len / 2);
470  x = L(2).x + r * cos(theta);
471  y = L(2).y + r * sin(theta);
472 
473  cx = (L(3).x + L(2).x + x) / 3;
474  cy = (L(3).y + L(2).y + y) / 3;
475 
476  L(2).x = x;
477  L(2).y = y;
478 
479  in1 = cx - L(3).x;
480  out1 = cy - L(3).y;
481  in2 = L(2).x - L(3).x;
482  out2 = L(2).y - L(3).y;
483  L(3).a = (out2 / in2 - out1 / in1) / (in2 - in1);
484  L(3).b = out1 / in1 - L(3).a * in1;
485  }
486  L(3).x = 0;
487  L(3).y = L(2).y;
488 
489  s->in_min_lin = exp(s->segments[1].x);
490  s->out_min_lin = exp(s->segments[1].y);
491 
492  for (i = 0; i < channels; i++) {
493  ChanParam *cp = &s->channels[i];
494 
495  if (cp->attack > 1.0 / sample_rate)
496  cp->attack = 1.0 - exp(-1.0 / (sample_rate * cp->attack));
497  else
498  cp->attack = 1.0;
499  if (cp->decay > 1.0 / sample_rate)
500  cp->decay = 1.0 - exp(-1.0 / (sample_rate * cp->decay));
501  else
502  cp->decay = 1.0;
503  cp->volume = pow(10.0, s->initial_volume / 20);
504  }
505 
506  s->delay_samples = s->delay * sample_rate;
507  if (s->delay_samples <= 0) {
509  return 0;
510  }
511 
513  if (!s->delay_frame) {
514  uninit(ctx);
515  return AVERROR(ENOMEM);
516  }
517 
518  s->delay_frame->format = outlink->format;
521 
522  err = av_frame_get_buffer(s->delay_frame, 32);
523  if (err)
524  return err;
525 
526  outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP;
527  s->compand = compand_delay;
528  return 0;
529 }
530 
531 static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
532 {
533  AVFilterContext *ctx = inlink->dst;
534  CompandContext *s = ctx->priv;
535 
536  return s->compand(ctx, frame);
537 }
538 
539 static int request_frame(AVFilterLink *outlink)
540 {
541  AVFilterContext *ctx = outlink->src;
542  CompandContext *s = ctx->priv;
543  int ret = 0;
544 
545  ret = ff_request_frame(ctx->inputs[0]);
546 
547  if (ret == AVERROR_EOF && !ctx->is_disabled && s->delay_count)
548  ret = compand_drain(outlink);
549 
550  return ret;
551 }
552 
553 static const AVFilterPad compand_inputs[] = {
554  {
555  .name = "default",
556  .type = AVMEDIA_TYPE_AUDIO,
557  .filter_frame = filter_frame,
558  },
559  { NULL }
560 };
561 
562 static const AVFilterPad compand_outputs[] = {
563  {
564  .name = "default",
565  .request_frame = request_frame,
566  .config_props = config_output,
567  .type = AVMEDIA_TYPE_AUDIO,
568  },
569  { NULL }
570 };
571 
572 
574  .name = "compand",
575  .description = NULL_IF_CONFIG_SMALL(
576  "Compress or expand audio dynamic range."),
577  .query_formats = query_formats,
578  .priv_size = sizeof(CompandContext),
579  .priv_class = &compand_class,
580  .init = init,
581  .uninit = uninit,
582  .inputs = compand_inputs,
583  .outputs = compand_outputs,
584 };