2 * Copyright (c) 2009 Rob Sykes <robs@users.sourceforge.net>
3 * Copyright (c) 2013 Paul B Mahol
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 #include "libavutil/opt.h"
29 typedef struct ChannelStats {
33 double sigma_x, sigma_x2;
34 double avg_sigma_x2, min_sigma_x2, max_sigma_x2;
37 double min_run, max_run;
38 double min_runs, max_runs;
39 double min_diff, max_diff;
43 uint64_t min_count, max_count;
48 typedef struct AudioStatsContext {
50 ChannelStats *chstats;
61 #define OFFSET(x) offsetof(AudioStatsContext, x)
62 #define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
64 static const AVOption astats_options[] = {
65 { "length", "set the window length", OFFSET(time_constant), AV_OPT_TYPE_DOUBLE, {.dbl=.05}, .01, 10, FLAGS },
66 { "metadata", "inject metadata in the filtergraph", OFFSET(metadata), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS },
67 { "reset", "recalculate stats after this many frames", OFFSET(reset_count), AV_OPT_TYPE_INT, {.i64=0}, 0, INT_MAX, FLAGS },
71 AVFILTER_DEFINE_CLASS(astats);
73 static int query_formats(AVFilterContext *ctx)
75 AVFilterFormats *formats;
76 AVFilterChannelLayouts *layouts;
77 static const enum AVSampleFormat sample_fmts[] = {
78 AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P,
79 AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_S32P,
80 AV_SAMPLE_FMT_S64, AV_SAMPLE_FMT_S64P,
81 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
82 AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_DBLP,
87 layouts = ff_all_channel_counts();
89 return AVERROR(ENOMEM);
90 ret = ff_set_common_channel_layouts(ctx, layouts);
94 formats = ff_make_format_list(sample_fmts);
96 return AVERROR(ENOMEM);
97 ret = ff_set_common_formats(ctx, formats);
101 formats = ff_all_samplerates();
103 return AVERROR(ENOMEM);
104 return ff_set_common_samplerates(ctx, formats);
107 static void reset_stats(AudioStatsContext *s)
111 for (c = 0; c < s->nb_channels; c++) {
112 ChannelStats *p = &s->chstats[c];
114 p->min = p->nmin = p->min_sigma_x2 = DBL_MAX;
115 p->max = p->nmax = p->max_sigma_x2 = DBL_MIN;
116 p->min_non_zero = DBL_MAX;
117 p->min_diff = DBL_MAX;
118 p->max_diff = DBL_MIN;
129 p->imask = 0xFFFFFFFFFFFFFFFF;
137 static int config_output(AVFilterLink *outlink)
139 AudioStatsContext *s = outlink->src->priv;
141 s->chstats = av_calloc(sizeof(*s->chstats), outlink->channels);
143 return AVERROR(ENOMEM);
144 s->nb_channels = outlink->channels;
145 s->mult = exp((-1 / s->time_constant / outlink->sample_rate));
146 s->tc_samples = 5 * s->time_constant * outlink->sample_rate + .5;
148 s->maxbitdepth = av_get_bytes_per_sample(outlink->format) * 8;
155 static void bit_depth(AudioStatsContext *s, uint64_t mask, uint64_t imask, AVRational *depth)
157 unsigned result = s->maxbitdepth;
159 mask = mask & (~imask);
161 for (; result && !(mask & 1); --result, mask >>= 1);
166 for (; result; --result, mask >>= 1)
171 static inline void update_stat(AudioStatsContext *s, ChannelStats *p, double d, double nd, int64_t i)
179 } else if (d == p->min) {
181 p->min_run = d == p->last ? p->min_run + 1 : 1;
182 } else if (p->last == p->min) {
183 p->min_runs += p->min_run * p->min_run;
186 if (d != 0 && FFABS(d) < p->min_non_zero)
187 p->min_non_zero = FFABS(d);
195 } else if (d == p->max) {
197 p->max_run = d == p->last ? p->max_run + 1 : 1;
198 } else if (p->last == p->max) {
199 p->max_runs += p->max_run * p->max_run;
203 p->zero_runs += FFSIGN(d) != FFSIGN(p->last_non_zero);
204 p->last_non_zero = d;
208 p->sigma_x2 += nd * nd;
209 p->avg_sigma_x2 = p->avg_sigma_x2 * s->mult + (1.0 - s->mult) * nd * nd;
210 p->min_diff = FFMIN(p->min_diff, fabs(d - p->last));
211 p->max_diff = FFMAX(p->max_diff, fabs(d - p->last));
212 p->diff1_sum += fabs(d - p->last);
213 p->diff1_sum_x2 += (d - p->last) * (d - p->last);
218 if (p->nb_samples >= s->tc_samples) {
219 p->max_sigma_x2 = FFMAX(p->max_sigma_x2, p->avg_sigma_x2);
220 p->min_sigma_x2 = FFMIN(p->min_sigma_x2, p->avg_sigma_x2);
225 static void set_meta(AVDictionary **metadata, int chan, const char *key,
226 const char *fmt, double val)
231 snprintf(value, sizeof(value), fmt, val);
233 snprintf(key2, sizeof(key2), "lavfi.astats.%d.%s", chan, key);
235 snprintf(key2, sizeof(key2), "lavfi.astats.%s", key);
236 av_dict_set(metadata, key2, value, 0);
239 #define LINEAR_TO_DB(x) (log10(x) * 20)
241 static void set_metadata(AudioStatsContext *s, AVDictionary **metadata)
243 uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0;
244 double min_runs = 0, max_runs = 0,
245 min = DBL_MAX, max = DBL_MIN, min_diff = DBL_MAX, max_diff = 0,
246 nmin = DBL_MAX, nmax = DBL_MIN,
252 min_sigma_x2 = DBL_MAX,
253 max_sigma_x2 = DBL_MIN;
257 for (c = 0; c < s->nb_channels; c++) {
258 ChannelStats *p = &s->chstats[c];
260 if (p->nb_samples < s->tc_samples)
261 p->min_sigma_x2 = p->max_sigma_x2 = p->sigma_x2 / p->nb_samples;
263 min = FFMIN(min, p->min);
264 max = FFMAX(max, p->max);
265 nmin = FFMIN(nmin, p->nmin);
266 nmax = FFMAX(nmax, p->nmax);
267 min_diff = FFMIN(min_diff, p->min_diff);
268 max_diff = FFMAX(max_diff, p->max_diff);
269 diff1_sum += p->diff1_sum;
270 diff1_sum_x2 += p->diff1_sum_x2;
271 min_sigma_x2 = FFMIN(min_sigma_x2, p->min_sigma_x2);
272 max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
273 sigma_x += p->sigma_x;
274 sigma_x2 += p->sigma_x2;
275 min_count += p->min_count;
276 max_count += p->max_count;
277 min_runs += p->min_runs;
278 max_runs += p->max_runs;
281 nb_samples += p->nb_samples;
282 if (fabs(p->sigma_x) > fabs(max_sigma_x))
283 max_sigma_x = p->sigma_x;
285 set_meta(metadata, c + 1, "DC_offset", "%f", p->sigma_x / p->nb_samples);
286 set_meta(metadata, c + 1, "Min_level", "%f", p->min);
287 set_meta(metadata, c + 1, "Max_level", "%f", p->max);
288 set_meta(metadata, c + 1, "Min_difference", "%f", p->min_diff);
289 set_meta(metadata, c + 1, "Max_difference", "%f", p->max_diff);
290 set_meta(metadata, c + 1, "Mean_difference", "%f", p->diff1_sum / (p->nb_samples - 1));
291 set_meta(metadata, c + 1, "RMS_difference", "%f", sqrt(p->diff1_sum_x2 / (p->nb_samples - 1)));
292 set_meta(metadata, c + 1, "Peak_level", "%f", LINEAR_TO_DB(FFMAX(-p->nmin, p->nmax)));
293 set_meta(metadata, c + 1, "RMS_level", "%f", LINEAR_TO_DB(sqrt(p->sigma_x2 / p->nb_samples)));
294 set_meta(metadata, c + 1, "RMS_peak", "%f", LINEAR_TO_DB(sqrt(p->max_sigma_x2)));
295 set_meta(metadata, c + 1, "RMS_trough", "%f", LINEAR_TO_DB(sqrt(p->min_sigma_x2)));
296 set_meta(metadata, c + 1, "Crest_factor", "%f", p->sigma_x2 ? FFMAX(-p->min, p->max) / sqrt(p->sigma_x2 / p->nb_samples) : 1);
297 set_meta(metadata, c + 1, "Flat_factor", "%f", LINEAR_TO_DB((p->min_runs + p->max_runs) / (p->min_count + p->max_count)));
298 set_meta(metadata, c + 1, "Peak_count", "%f", (float)(p->min_count + p->max_count));
299 bit_depth(s, p->mask, p->imask, &depth);
300 set_meta(metadata, c + 1, "Bit_depth", "%f", depth.num);
301 set_meta(metadata, c + 1, "Bit_depth2", "%f", depth.den);
302 set_meta(metadata, c + 1, "Dynamic_range", "%f", LINEAR_TO_DB(2 * FFMAX(FFABS(p->min), FFABS(p->max))/ p->min_non_zero));
303 set_meta(metadata, c + 1, "Zero_crossings", "%f", p->zero_runs);
304 set_meta(metadata, c + 1, "Zero_crossings_rate", "%f", p->zero_runs/(double)p->nb_samples);
307 set_meta(metadata, 0, "Overall.DC_offset", "%f", max_sigma_x / (nb_samples / s->nb_channels));
308 set_meta(metadata, 0, "Overall.Min_level", "%f", min);
309 set_meta(metadata, 0, "Overall.Max_level", "%f", max);
310 set_meta(metadata, 0, "Overall.Min_difference", "%f", min_diff);
311 set_meta(metadata, 0, "Overall.Max_difference", "%f", max_diff);
312 set_meta(metadata, 0, "Overall.Mean_difference", "%f", diff1_sum / (nb_samples - s->nb_channels));
313 set_meta(metadata, 0, "Overall.RMS_difference", "%f", sqrt(diff1_sum_x2 / (nb_samples - s->nb_channels)));
314 set_meta(metadata, 0, "Overall.Peak_level", "%f", LINEAR_TO_DB(FFMAX(-nmin, nmax)));
315 set_meta(metadata, 0, "Overall.RMS_level", "%f", LINEAR_TO_DB(sqrt(sigma_x2 / nb_samples)));
316 set_meta(metadata, 0, "Overall.RMS_peak", "%f", LINEAR_TO_DB(sqrt(max_sigma_x2)));
317 set_meta(metadata, 0, "Overall.RMS_trough", "%f", LINEAR_TO_DB(sqrt(min_sigma_x2)));
318 set_meta(metadata, 0, "Overall.Flat_factor", "%f", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
319 set_meta(metadata, 0, "Overall.Peak_count", "%f", (float)(min_count + max_count) / (double)s->nb_channels);
320 bit_depth(s, mask, imask, &depth);
321 set_meta(metadata, 0, "Overall.Bit_depth", "%f", depth.num);
322 set_meta(metadata, 0, "Overall.Bit_depth2", "%f", depth.den);
323 set_meta(metadata, 0, "Overall.Number_of_samples", "%f", nb_samples / s->nb_channels);
326 static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
328 AudioStatsContext *s = inlink->dst->priv;
329 AVDictionary **metadata = &buf->metadata;
330 const int channels = s->nb_channels;
333 if (s->reset_count > 0) {
334 if (s->nb_frames >= s->reset_count) {
341 switch (inlink->format) {
342 case AV_SAMPLE_FMT_DBLP:
343 for (c = 0; c < channels; c++) {
344 ChannelStats *p = &s->chstats[c];
345 const double *src = (const double *)buf->extended_data[c];
347 for (i = 0; i < buf->nb_samples; i++, src++)
348 update_stat(s, p, *src, *src, llrint(*src * (UINT64_C(1) << 63)));
351 case AV_SAMPLE_FMT_DBL: {
352 const double *src = (const double *)buf->extended_data[0];
354 for (i = 0; i < buf->nb_samples; i++) {
355 for (c = 0; c < channels; c++, src++)
356 update_stat(s, &s->chstats[c], *src, *src, llrint(*src * (UINT64_C(1) << 63)));
359 case AV_SAMPLE_FMT_FLTP:
360 for (c = 0; c < channels; c++) {
361 ChannelStats *p = &s->chstats[c];
362 const float *src = (const float *)buf->extended_data[c];
364 for (i = 0; i < buf->nb_samples; i++, src++)
365 update_stat(s, p, *src, *src, llrint(*src * (UINT64_C(1) << 31)));
368 case AV_SAMPLE_FMT_FLT: {
369 const float *src = (const float *)buf->extended_data[0];
371 for (i = 0; i < buf->nb_samples; i++) {
372 for (c = 0; c < channels; c++, src++)
373 update_stat(s, &s->chstats[c], *src, *src, llrint(*src * (UINT64_C(1) << 31)));
376 case AV_SAMPLE_FMT_S64P:
377 for (c = 0; c < channels; c++) {
378 ChannelStats *p = &s->chstats[c];
379 const int64_t *src = (const int64_t *)buf->extended_data[c];
381 for (i = 0; i < buf->nb_samples; i++, src++)
382 update_stat(s, p, *src, *src / (double)INT64_MAX, *src);
385 case AV_SAMPLE_FMT_S64: {
386 const int64_t *src = (const int64_t *)buf->extended_data[0];
388 for (i = 0; i < buf->nb_samples; i++) {
389 for (c = 0; c < channels; c++, src++)
390 update_stat(s, &s->chstats[c], *src, *src / (double)INT64_MAX, *src);
393 case AV_SAMPLE_FMT_S32P:
394 for (c = 0; c < channels; c++) {
395 ChannelStats *p = &s->chstats[c];
396 const int32_t *src = (const int32_t *)buf->extended_data[c];
398 for (i = 0; i < buf->nb_samples; i++, src++)
399 update_stat(s, p, *src, *src / (double)INT32_MAX, *src);
402 case AV_SAMPLE_FMT_S32: {
403 const int32_t *src = (const int32_t *)buf->extended_data[0];
405 for (i = 0; i < buf->nb_samples; i++) {
406 for (c = 0; c < channels; c++, src++)
407 update_stat(s, &s->chstats[c], *src, *src / (double)INT32_MAX, *src);
410 case AV_SAMPLE_FMT_S16P:
411 for (c = 0; c < channels; c++) {
412 ChannelStats *p = &s->chstats[c];
413 const int16_t *src = (const int16_t *)buf->extended_data[c];
415 for (i = 0; i < buf->nb_samples; i++, src++)
416 update_stat(s, p, *src, *src / (double)INT16_MAX, *src);
419 case AV_SAMPLE_FMT_S16: {
420 const int16_t *src = (const int16_t *)buf->extended_data[0];
422 for (i = 0; i < buf->nb_samples; i++) {
423 for (c = 0; c < channels; c++, src++)
424 update_stat(s, &s->chstats[c], *src, *src / (double)INT16_MAX, *src);
430 set_metadata(s, metadata);
432 return ff_filter_frame(inlink->dst->outputs[0], buf);
435 static void print_stats(AVFilterContext *ctx)
437 AudioStatsContext *s = ctx->priv;
438 uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0;
439 double min_runs = 0, max_runs = 0,
440 min = DBL_MAX, max = DBL_MIN, min_diff = DBL_MAX, max_diff = 0,
441 nmin = DBL_MAX, nmax = DBL_MIN,
447 min_sigma_x2 = DBL_MAX,
448 max_sigma_x2 = DBL_MIN;
452 for (c = 0; c < s->nb_channels; c++) {
453 ChannelStats *p = &s->chstats[c];
455 if (p->nb_samples < s->tc_samples)
456 p->min_sigma_x2 = p->max_sigma_x2 = p->sigma_x2 / p->nb_samples;
458 min = FFMIN(min, p->min);
459 max = FFMAX(max, p->max);
460 nmin = FFMIN(nmin, p->nmin);
461 nmax = FFMAX(nmax, p->nmax);
462 min_diff = FFMIN(min_diff, p->min_diff);
463 max_diff = FFMAX(max_diff, p->max_diff);
464 diff1_sum_x2 += p->diff1_sum_x2;
465 diff1_sum += p->diff1_sum;
466 min_sigma_x2 = FFMIN(min_sigma_x2, p->min_sigma_x2);
467 max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
468 sigma_x += p->sigma_x;
469 sigma_x2 += p->sigma_x2;
470 min_count += p->min_count;
471 max_count += p->max_count;
472 min_runs += p->min_runs;
473 max_runs += p->max_runs;
476 nb_samples += p->nb_samples;
477 if (fabs(p->sigma_x) > fabs(max_sigma_x))
478 max_sigma_x = p->sigma_x;
480 av_log(ctx, AV_LOG_INFO, "Channel: %d\n", c + 1);
481 av_log(ctx, AV_LOG_INFO, "DC offset: %f\n", p->sigma_x / p->nb_samples);
482 av_log(ctx, AV_LOG_INFO, "Min level: %f\n", p->min);
483 av_log(ctx, AV_LOG_INFO, "Max level: %f\n", p->max);
484 av_log(ctx, AV_LOG_INFO, "Min difference: %f\n", p->min_diff);
485 av_log(ctx, AV_LOG_INFO, "Max difference: %f\n", p->max_diff);
486 av_log(ctx, AV_LOG_INFO, "Mean difference: %f\n", p->diff1_sum / (p->nb_samples - 1));
487 av_log(ctx, AV_LOG_INFO, "RMS difference: %f\n", sqrt(p->diff1_sum_x2 / (p->nb_samples - 1)));
488 av_log(ctx, AV_LOG_INFO, "Peak level dB: %f\n", LINEAR_TO_DB(FFMAX(-p->nmin, p->nmax)));
489 av_log(ctx, AV_LOG_INFO, "RMS level dB: %f\n", LINEAR_TO_DB(sqrt(p->sigma_x2 / p->nb_samples)));
490 av_log(ctx, AV_LOG_INFO, "RMS peak dB: %f\n", LINEAR_TO_DB(sqrt(p->max_sigma_x2)));
491 if (p->min_sigma_x2 != 1)
492 av_log(ctx, AV_LOG_INFO, "RMS trough dB: %f\n",LINEAR_TO_DB(sqrt(p->min_sigma_x2)));
493 av_log(ctx, AV_LOG_INFO, "Crest factor: %f\n", p->sigma_x2 ? FFMAX(-p->nmin, p->nmax) / sqrt(p->sigma_x2 / p->nb_samples) : 1);
494 av_log(ctx, AV_LOG_INFO, "Flat factor: %f\n", LINEAR_TO_DB((p->min_runs + p->max_runs) / (p->min_count + p->max_count)));
495 av_log(ctx, AV_LOG_INFO, "Peak count: %"PRId64"\n", p->min_count + p->max_count);
496 bit_depth(s, p->mask, p->imask, &depth);
497 av_log(ctx, AV_LOG_INFO, "Bit depth: %u/%u\n", depth.num, depth.den);
498 av_log(ctx, AV_LOG_INFO, "Dynamic range: %f\n", LINEAR_TO_DB(2 * FFMAX(FFABS(p->min), FFABS(p->max))/ p->min_non_zero));
499 av_log(ctx, AV_LOG_INFO, "Zero crossings: %"PRId64"\n", p->zero_runs);
500 av_log(ctx, AV_LOG_INFO, "Zero crossings rate: %f\n", p->zero_runs/(double)p->nb_samples);
503 av_log(ctx, AV_LOG_INFO, "Overall\n");
504 av_log(ctx, AV_LOG_INFO, "DC offset: %f\n", max_sigma_x / (nb_samples / s->nb_channels));
505 av_log(ctx, AV_LOG_INFO, "Min level: %f\n", min);
506 av_log(ctx, AV_LOG_INFO, "Max level: %f\n", max);
507 av_log(ctx, AV_LOG_INFO, "Min difference: %f\n", min_diff);
508 av_log(ctx, AV_LOG_INFO, "Max difference: %f\n", max_diff);
509 av_log(ctx, AV_LOG_INFO, "Mean difference: %f\n", diff1_sum / (nb_samples - s->nb_channels));
510 av_log(ctx, AV_LOG_INFO, "RMS difference: %f\n", sqrt(diff1_sum_x2 / (nb_samples - s->nb_channels)));
511 av_log(ctx, AV_LOG_INFO, "Peak level dB: %f\n", LINEAR_TO_DB(FFMAX(-nmin, nmax)));
512 av_log(ctx, AV_LOG_INFO, "RMS level dB: %f\n", LINEAR_TO_DB(sqrt(sigma_x2 / nb_samples)));
513 av_log(ctx, AV_LOG_INFO, "RMS peak dB: %f\n", LINEAR_TO_DB(sqrt(max_sigma_x2)));
514 if (min_sigma_x2 != 1)
515 av_log(ctx, AV_LOG_INFO, "RMS trough dB: %f\n", LINEAR_TO_DB(sqrt(min_sigma_x2)));
516 av_log(ctx, AV_LOG_INFO, "Flat factor: %f\n", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
517 av_log(ctx, AV_LOG_INFO, "Peak count: %f\n", (min_count + max_count) / (double)s->nb_channels);
518 bit_depth(s, mask, imask, &depth);
519 av_log(ctx, AV_LOG_INFO, "Bit depth: %u/%u\n", depth.num, depth.den);
520 av_log(ctx, AV_LOG_INFO, "Number of samples: %"PRId64"\n", nb_samples / s->nb_channels);
523 static av_cold void uninit(AVFilterContext *ctx)
525 AudioStatsContext *s = ctx->priv;
529 av_freep(&s->chstats);
532 static const AVFilterPad astats_inputs[] = {
535 .type = AVMEDIA_TYPE_AUDIO,
536 .filter_frame = filter_frame,
541 static const AVFilterPad astats_outputs[] = {
544 .type = AVMEDIA_TYPE_AUDIO,
545 .config_props = config_output,
550 AVFilter ff_af_astats = {
552 .description = NULL_IF_CONFIG_SMALL("Show time domain statistics about audio frames."),
553 .query_formats = query_formats,
554 .priv_size = sizeof(AudioStatsContext),
555 .priv_class = &astats_class,
557 .inputs = astats_inputs,
558 .outputs = astats_outputs,