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 {
31 double sigma_x, sigma_x2;
32 double avg_sigma_x2, min_sigma_x2, max_sigma_x2;
35 double min_run, max_run;
36 double min_runs, max_runs;
37 double min_diff, max_diff;
41 uint64_t min_count, max_count;
45 typedef struct AudioStatsContext {
47 ChannelStats *chstats;
58 #define OFFSET(x) offsetof(AudioStatsContext, x)
59 #define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
61 static const AVOption astats_options[] = {
62 { "length", "set the window length", OFFSET(time_constant), AV_OPT_TYPE_DOUBLE, {.dbl=.05}, .01, 10, FLAGS },
63 { "metadata", "inject metadata in the filtergraph", OFFSET(metadata), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS },
64 { "reset", "recalculate stats after this many frames", OFFSET(reset_count), AV_OPT_TYPE_INT, {.i64=0}, 0, INT_MAX, FLAGS },
68 AVFILTER_DEFINE_CLASS(astats);
70 static int query_formats(AVFilterContext *ctx)
72 AVFilterFormats *formats;
73 AVFilterChannelLayouts *layouts;
74 static const enum AVSampleFormat sample_fmts[] = {
75 AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P,
76 AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_S32P,
77 AV_SAMPLE_FMT_S64, AV_SAMPLE_FMT_S64P,
78 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
79 AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_DBLP,
84 layouts = ff_all_channel_counts();
86 return AVERROR(ENOMEM);
87 ret = ff_set_common_channel_layouts(ctx, layouts);
91 formats = ff_make_format_list(sample_fmts);
93 return AVERROR(ENOMEM);
94 ret = ff_set_common_formats(ctx, formats);
98 formats = ff_all_samplerates();
100 return AVERROR(ENOMEM);
101 return ff_set_common_samplerates(ctx, formats);
104 static void reset_stats(AudioStatsContext *s)
108 for (c = 0; c < s->nb_channels; c++) {
109 ChannelStats *p = &s->chstats[c];
111 p->min = p->nmin = p->min_sigma_x2 = DBL_MAX;
112 p->max = p->nmax = p->max_sigma_x2 = DBL_MIN;
113 p->min_diff = DBL_MAX;
114 p->max_diff = DBL_MIN;
127 p->imask = 0xFFFFFFFFFFFFFFFF;
134 static int config_output(AVFilterLink *outlink)
136 AudioStatsContext *s = outlink->src->priv;
138 s->chstats = av_calloc(sizeof(*s->chstats), outlink->channels);
140 return AVERROR(ENOMEM);
141 s->nb_channels = outlink->channels;
142 s->mult = exp((-1 / s->time_constant / outlink->sample_rate));
143 s->tc_samples = 5 * s->time_constant * outlink->sample_rate + .5;
145 s->maxbitdepth = av_get_bytes_per_sample(outlink->format) * 8;
152 static void bit_depth(AudioStatsContext *s, uint64_t mask, uint64_t imask, AVRational *depth)
154 unsigned result = s->maxbitdepth;
156 mask = mask & (~imask);
158 for (; result && !(mask & 1); --result, mask >>= 1);
163 for (; result; --result, mask >>= 1)
168 static inline void update_stat(AudioStatsContext *s, ChannelStats *p, double d, double nd, int64_t i)
176 } else if (d == p->min) {
178 p->min_run = d == p->last ? p->min_run + 1 : 1;
179 } else if (p->last == p->min) {
180 p->min_runs += p->min_run * p->min_run;
189 } else if (d == p->max) {
191 p->max_run = d == p->last ? p->max_run + 1 : 1;
192 } else if (p->last == p->max) {
193 p->max_runs += p->max_run * p->max_run;
197 p->sigma_x2 += nd * nd;
198 p->avg_sigma_x2 = p->avg_sigma_x2 * s->mult + (1.0 - s->mult) * nd * nd;
199 p->min_diff = FFMIN(p->min_diff, fabs(d - p->last));
200 p->max_diff = FFMAX(p->max_diff, fabs(d - p->last));
201 p->diff1_sum += fabs(d - p->last);
202 p->diff1_sum_x2 += (d - p->last) * (d - p->last);
207 if (p->nb_samples >= s->tc_samples) {
208 p->max_sigma_x2 = FFMAX(p->max_sigma_x2, p->avg_sigma_x2);
209 p->min_sigma_x2 = FFMIN(p->min_sigma_x2, p->avg_sigma_x2);
214 static void set_meta(AVDictionary **metadata, int chan, const char *key,
215 const char *fmt, double val)
220 snprintf(value, sizeof(value), fmt, val);
222 snprintf(key2, sizeof(key2), "lavfi.astats.%d.%s", chan, key);
224 snprintf(key2, sizeof(key2), "lavfi.astats.%s", key);
225 av_dict_set(metadata, key2, value, 0);
228 #define LINEAR_TO_DB(x) (log10(x) * 20)
230 static void set_metadata(AudioStatsContext *s, AVDictionary **metadata)
232 uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0;
233 double min_runs = 0, max_runs = 0,
234 min = DBL_MAX, max = DBL_MIN, min_diff = DBL_MAX, max_diff = 0,
235 nmin = DBL_MAX, nmax = DBL_MIN,
241 min_sigma_x2 = DBL_MAX,
242 max_sigma_x2 = DBL_MIN;
246 for (c = 0; c < s->nb_channels; c++) {
247 ChannelStats *p = &s->chstats[c];
249 if (p->nb_samples < s->tc_samples)
250 p->min_sigma_x2 = p->max_sigma_x2 = p->sigma_x2 / p->nb_samples;
252 min = FFMIN(min, p->min);
253 max = FFMAX(max, p->max);
254 nmin = FFMIN(nmin, p->nmin);
255 nmax = FFMAX(nmax, p->nmax);
256 min_diff = FFMIN(min_diff, p->min_diff);
257 max_diff = FFMAX(max_diff, p->max_diff);
258 diff1_sum += p->diff1_sum;
259 diff1_sum_x2 += p->diff1_sum_x2;
260 min_sigma_x2 = FFMIN(min_sigma_x2, p->min_sigma_x2);
261 max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
262 sigma_x += p->sigma_x;
263 sigma_x2 += p->sigma_x2;
264 min_count += p->min_count;
265 max_count += p->max_count;
266 min_runs += p->min_runs;
267 max_runs += p->max_runs;
270 nb_samples += p->nb_samples;
271 if (fabs(p->sigma_x) > fabs(max_sigma_x))
272 max_sigma_x = p->sigma_x;
274 set_meta(metadata, c + 1, "DC_offset", "%f", p->sigma_x / p->nb_samples);
275 set_meta(metadata, c + 1, "Min_level", "%f", p->min);
276 set_meta(metadata, c + 1, "Max_level", "%f", p->max);
277 set_meta(metadata, c + 1, "Min_difference", "%f", p->min_diff);
278 set_meta(metadata, c + 1, "Max_difference", "%f", p->max_diff);
279 set_meta(metadata, c + 1, "Mean_difference", "%f", p->diff1_sum / (p->nb_samples - 1));
280 set_meta(metadata, c + 1, "RMS_difference", "%f", sqrt(p->diff1_sum_x2 / (p->nb_samples - 1)));
281 set_meta(metadata, c + 1, "Peak_level", "%f", LINEAR_TO_DB(FFMAX(-p->nmin, p->nmax)));
282 set_meta(metadata, c + 1, "RMS_level", "%f", LINEAR_TO_DB(sqrt(p->sigma_x2 / p->nb_samples)));
283 set_meta(metadata, c + 1, "RMS_peak", "%f", LINEAR_TO_DB(sqrt(p->max_sigma_x2)));
284 set_meta(metadata, c + 1, "RMS_trough", "%f", LINEAR_TO_DB(sqrt(p->min_sigma_x2)));
285 set_meta(metadata, c + 1, "Crest_factor", "%f", p->sigma_x2 ? FFMAX(-p->min, p->max) / sqrt(p->sigma_x2 / p->nb_samples) : 1);
286 set_meta(metadata, c + 1, "Flat_factor", "%f", LINEAR_TO_DB((p->min_runs + p->max_runs) / (p->min_count + p->max_count)));
287 set_meta(metadata, c + 1, "Peak_count", "%f", (float)(p->min_count + p->max_count));
288 bit_depth(s, p->mask, p->imask, &depth);
289 set_meta(metadata, c + 1, "Bit_depth", "%f", depth.num);
290 set_meta(metadata, c + 1, "Bit_depth2", "%f", depth.den);
293 set_meta(metadata, 0, "Overall.DC_offset", "%f", max_sigma_x / (nb_samples / s->nb_channels));
294 set_meta(metadata, 0, "Overall.Min_level", "%f", min);
295 set_meta(metadata, 0, "Overall.Max_level", "%f", max);
296 set_meta(metadata, 0, "Overall.Min_difference", "%f", min_diff);
297 set_meta(metadata, 0, "Overall.Max_difference", "%f", max_diff);
298 set_meta(metadata, 0, "Overall.Mean_difference", "%f", diff1_sum / (nb_samples - s->nb_channels));
299 set_meta(metadata, 0, "Overall.RMS_difference", "%f", sqrt(diff1_sum_x2 / (nb_samples - s->nb_channels)));
300 set_meta(metadata, 0, "Overall.Peak_level", "%f", LINEAR_TO_DB(FFMAX(-nmin, nmax)));
301 set_meta(metadata, 0, "Overall.RMS_level", "%f", LINEAR_TO_DB(sqrt(sigma_x2 / nb_samples)));
302 set_meta(metadata, 0, "Overall.RMS_peak", "%f", LINEAR_TO_DB(sqrt(max_sigma_x2)));
303 set_meta(metadata, 0, "Overall.RMS_trough", "%f", LINEAR_TO_DB(sqrt(min_sigma_x2)));
304 set_meta(metadata, 0, "Overall.Flat_factor", "%f", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
305 set_meta(metadata, 0, "Overall.Peak_count", "%f", (float)(min_count + max_count) / (double)s->nb_channels);
306 bit_depth(s, mask, imask, &depth);
307 set_meta(metadata, 0, "Overall.Bit_depth", "%f", depth.num);
308 set_meta(metadata, 0, "Overall.Bit_depth2", "%f", depth.den);
309 set_meta(metadata, 0, "Overall.Number_of_samples", "%f", nb_samples / s->nb_channels);
312 static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
314 AudioStatsContext *s = inlink->dst->priv;
315 AVDictionary **metadata = &buf->metadata;
316 const int channels = s->nb_channels;
319 if (s->reset_count > 0) {
320 if (s->nb_frames >= s->reset_count) {
327 switch (inlink->format) {
328 case AV_SAMPLE_FMT_DBLP:
329 for (c = 0; c < channels; c++) {
330 ChannelStats *p = &s->chstats[c];
331 const double *src = (const double *)buf->extended_data[c];
333 for (i = 0; i < buf->nb_samples; i++, src++)
334 update_stat(s, p, *src, *src, llrint(*src * (UINT64_C(1) << 63)));
337 case AV_SAMPLE_FMT_DBL: {
338 const double *src = (const double *)buf->extended_data[0];
340 for (i = 0; i < buf->nb_samples; i++) {
341 for (c = 0; c < channels; c++, src++)
342 update_stat(s, &s->chstats[c], *src, *src, llrint(*src * (UINT64_C(1) << 63)));
345 case AV_SAMPLE_FMT_FLTP:
346 for (c = 0; c < channels; c++) {
347 ChannelStats *p = &s->chstats[c];
348 const float *src = (const float *)buf->extended_data[c];
350 for (i = 0; i < buf->nb_samples; i++, src++)
351 update_stat(s, p, *src, *src, llrint(*src * (UINT64_C(1) << 31)));
354 case AV_SAMPLE_FMT_FLT: {
355 const float *src = (const float *)buf->extended_data[0];
357 for (i = 0; i < buf->nb_samples; i++) {
358 for (c = 0; c < channels; c++, src++)
359 update_stat(s, &s->chstats[c], *src, *src, llrint(*src * (UINT64_C(1) << 31)));
362 case AV_SAMPLE_FMT_S64P:
363 for (c = 0; c < channels; c++) {
364 ChannelStats *p = &s->chstats[c];
365 const int64_t *src = (const int64_t *)buf->extended_data[c];
367 for (i = 0; i < buf->nb_samples; i++, src++)
368 update_stat(s, p, *src, *src / (double)INT64_MAX, *src);
371 case AV_SAMPLE_FMT_S64: {
372 const int64_t *src = (const int64_t *)buf->extended_data[0];
374 for (i = 0; i < buf->nb_samples; i++) {
375 for (c = 0; c < channels; c++, src++)
376 update_stat(s, &s->chstats[c], *src, *src / (double)INT64_MAX, *src);
379 case AV_SAMPLE_FMT_S32P:
380 for (c = 0; c < channels; c++) {
381 ChannelStats *p = &s->chstats[c];
382 const int32_t *src = (const int32_t *)buf->extended_data[c];
384 for (i = 0; i < buf->nb_samples; i++, src++)
385 update_stat(s, p, *src, *src / (double)INT32_MAX, *src);
388 case AV_SAMPLE_FMT_S32: {
389 const int32_t *src = (const int32_t *)buf->extended_data[0];
391 for (i = 0; i < buf->nb_samples; i++) {
392 for (c = 0; c < channels; c++, src++)
393 update_stat(s, &s->chstats[c], *src, *src / (double)INT32_MAX, *src);
396 case AV_SAMPLE_FMT_S16P:
397 for (c = 0; c < channels; c++) {
398 ChannelStats *p = &s->chstats[c];
399 const int16_t *src = (const int16_t *)buf->extended_data[c];
401 for (i = 0; i < buf->nb_samples; i++, src++)
402 update_stat(s, p, *src, *src / (double)INT16_MAX, *src);
405 case AV_SAMPLE_FMT_S16: {
406 const int16_t *src = (const int16_t *)buf->extended_data[0];
408 for (i = 0; i < buf->nb_samples; i++) {
409 for (c = 0; c < channels; c++, src++)
410 update_stat(s, &s->chstats[c], *src, *src / (double)INT16_MAX, *src);
416 set_metadata(s, metadata);
418 return ff_filter_frame(inlink->dst->outputs[0], buf);
421 static void print_stats(AVFilterContext *ctx)
423 AudioStatsContext *s = ctx->priv;
424 uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0;
425 double min_runs = 0, max_runs = 0,
426 min = DBL_MAX, max = DBL_MIN, min_diff = DBL_MAX, max_diff = 0,
427 nmin = DBL_MAX, nmax = DBL_MIN,
433 min_sigma_x2 = DBL_MAX,
434 max_sigma_x2 = DBL_MIN;
438 for (c = 0; c < s->nb_channels; c++) {
439 ChannelStats *p = &s->chstats[c];
441 if (p->nb_samples < s->tc_samples)
442 p->min_sigma_x2 = p->max_sigma_x2 = p->sigma_x2 / p->nb_samples;
444 min = FFMIN(min, p->min);
445 max = FFMAX(max, p->max);
446 nmin = FFMIN(nmin, p->nmin);
447 nmax = FFMAX(nmax, p->nmax);
448 min_diff = FFMIN(min_diff, p->min_diff);
449 max_diff = FFMAX(max_diff, p->max_diff);
450 diff1_sum_x2 += p->diff1_sum_x2;
451 diff1_sum += p->diff1_sum;
452 min_sigma_x2 = FFMIN(min_sigma_x2, p->min_sigma_x2);
453 max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
454 sigma_x += p->sigma_x;
455 sigma_x2 += p->sigma_x2;
456 min_count += p->min_count;
457 max_count += p->max_count;
458 min_runs += p->min_runs;
459 max_runs += p->max_runs;
462 nb_samples += p->nb_samples;
463 if (fabs(p->sigma_x) > fabs(max_sigma_x))
464 max_sigma_x = p->sigma_x;
466 av_log(ctx, AV_LOG_INFO, "Channel: %d\n", c + 1);
467 av_log(ctx, AV_LOG_INFO, "DC offset: %f\n", p->sigma_x / p->nb_samples);
468 av_log(ctx, AV_LOG_INFO, "Min level: %f\n", p->min);
469 av_log(ctx, AV_LOG_INFO, "Max level: %f\n", p->max);
470 av_log(ctx, AV_LOG_INFO, "Min difference: %f\n", p->min_diff);
471 av_log(ctx, AV_LOG_INFO, "Max difference: %f\n", p->max_diff);
472 av_log(ctx, AV_LOG_INFO, "Mean difference: %f\n", p->diff1_sum / (p->nb_samples - 1));
473 av_log(ctx, AV_LOG_INFO, "RMS difference: %f\n", sqrt(p->diff1_sum_x2 / (p->nb_samples - 1)));
474 av_log(ctx, AV_LOG_INFO, "Peak level dB: %f\n", LINEAR_TO_DB(FFMAX(-p->nmin, p->nmax)));
475 av_log(ctx, AV_LOG_INFO, "RMS level dB: %f\n", LINEAR_TO_DB(sqrt(p->sigma_x2 / p->nb_samples)));
476 av_log(ctx, AV_LOG_INFO, "RMS peak dB: %f\n", LINEAR_TO_DB(sqrt(p->max_sigma_x2)));
477 if (p->min_sigma_x2 != 1)
478 av_log(ctx, AV_LOG_INFO, "RMS trough dB: %f\n",LINEAR_TO_DB(sqrt(p->min_sigma_x2)));
479 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);
480 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)));
481 av_log(ctx, AV_LOG_INFO, "Peak count: %"PRId64"\n", p->min_count + p->max_count);
482 bit_depth(s, p->mask, p->imask, &depth);
483 av_log(ctx, AV_LOG_INFO, "Bit depth: %u/%u\n", depth.num, depth.den);
486 av_log(ctx, AV_LOG_INFO, "Overall\n");
487 av_log(ctx, AV_LOG_INFO, "DC offset: %f\n", max_sigma_x / (nb_samples / s->nb_channels));
488 av_log(ctx, AV_LOG_INFO, "Min level: %f\n", min);
489 av_log(ctx, AV_LOG_INFO, "Max level: %f\n", max);
490 av_log(ctx, AV_LOG_INFO, "Min difference: %f\n", min_diff);
491 av_log(ctx, AV_LOG_INFO, "Max difference: %f\n", max_diff);
492 av_log(ctx, AV_LOG_INFO, "Mean difference: %f\n", diff1_sum / (nb_samples - s->nb_channels));
493 av_log(ctx, AV_LOG_INFO, "RMS difference: %f\n", sqrt(diff1_sum_x2 / (nb_samples - s->nb_channels)));
494 av_log(ctx, AV_LOG_INFO, "Peak level dB: %f\n", LINEAR_TO_DB(FFMAX(-nmin, nmax)));
495 av_log(ctx, AV_LOG_INFO, "RMS level dB: %f\n", LINEAR_TO_DB(sqrt(sigma_x2 / nb_samples)));
496 av_log(ctx, AV_LOG_INFO, "RMS peak dB: %f\n", LINEAR_TO_DB(sqrt(max_sigma_x2)));
497 if (min_sigma_x2 != 1)
498 av_log(ctx, AV_LOG_INFO, "RMS trough dB: %f\n", LINEAR_TO_DB(sqrt(min_sigma_x2)));
499 av_log(ctx, AV_LOG_INFO, "Flat factor: %f\n", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
500 av_log(ctx, AV_LOG_INFO, "Peak count: %f\n", (min_count + max_count) / (double)s->nb_channels);
501 bit_depth(s, mask, imask, &depth);
502 av_log(ctx, AV_LOG_INFO, "Bit depth: %u/%u\n", depth.num, depth.den);
503 av_log(ctx, AV_LOG_INFO, "Number of samples: %"PRId64"\n", nb_samples / s->nb_channels);
506 static av_cold void uninit(AVFilterContext *ctx)
508 AudioStatsContext *s = ctx->priv;
512 av_freep(&s->chstats);
515 static const AVFilterPad astats_inputs[] = {
518 .type = AVMEDIA_TYPE_AUDIO,
519 .filter_frame = filter_frame,
524 static const AVFilterPad astats_outputs[] = {
527 .type = AVMEDIA_TYPE_AUDIO,
528 .config_props = config_output,
533 AVFilter ff_af_astats = {
535 .description = NULL_IF_CONFIG_SMALL("Show time domain statistics about audio frames."),
536 .query_formats = query_formats,
537 .priv_size = sizeof(AudioStatsContext),
538 .priv_class = &astats_class,
540 .inputs = astats_inputs,
541 .outputs = astats_outputs,