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 {
32 double sigma_x, sigma_x2;
33 double avg_sigma_x2, min_sigma_x2, max_sigma_x2;
36 double min_run, max_run;
37 double min_runs, max_runs;
38 double min_diff, max_diff;
42 uint64_t min_count, max_count;
46 typedef struct AudioStatsContext {
48 ChannelStats *chstats;
59 #define OFFSET(x) offsetof(AudioStatsContext, x)
60 #define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
62 static const AVOption astats_options[] = {
63 { "length", "set the window length", OFFSET(time_constant), AV_OPT_TYPE_DOUBLE, {.dbl=.05}, .01, 10, FLAGS },
64 { "metadata", "inject metadata in the filtergraph", OFFSET(metadata), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS },
65 { "reset", "recalculate stats after this many frames", OFFSET(reset_count), AV_OPT_TYPE_INT, {.i64=0}, 0, INT_MAX, FLAGS },
69 AVFILTER_DEFINE_CLASS(astats);
71 static int query_formats(AVFilterContext *ctx)
73 AVFilterFormats *formats;
74 AVFilterChannelLayouts *layouts;
75 static const enum AVSampleFormat sample_fmts[] = {
76 AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P,
77 AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_S32P,
78 AV_SAMPLE_FMT_S64, AV_SAMPLE_FMT_S64P,
79 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
80 AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_DBLP,
85 layouts = ff_all_channel_counts();
87 return AVERROR(ENOMEM);
88 ret = ff_set_common_channel_layouts(ctx, layouts);
92 formats = ff_make_format_list(sample_fmts);
94 return AVERROR(ENOMEM);
95 ret = ff_set_common_formats(ctx, formats);
99 formats = ff_all_samplerates();
101 return AVERROR(ENOMEM);
102 return ff_set_common_samplerates(ctx, formats);
105 static void reset_stats(AudioStatsContext *s)
109 for (c = 0; c < s->nb_channels; c++) {
110 ChannelStats *p = &s->chstats[c];
112 p->min = p->nmin = p->min_sigma_x2 = DBL_MAX;
113 p->max = p->nmax = p->max_sigma_x2 = DBL_MIN;
114 p->min_non_zero = DBL_MAX;
115 p->min_diff = DBL_MAX;
116 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;
183 if (d != 0 && FFABS(d) < p->min_non_zero)
184 p->min_non_zero = FFABS(d);
192 } else if (d == p->max) {
194 p->max_run = d == p->last ? p->max_run + 1 : 1;
195 } else if (p->last == p->max) {
196 p->max_runs += p->max_run * p->max_run;
200 p->sigma_x2 += nd * nd;
201 p->avg_sigma_x2 = p->avg_sigma_x2 * s->mult + (1.0 - s->mult) * nd * nd;
202 p->min_diff = FFMIN(p->min_diff, fabs(d - p->last));
203 p->max_diff = FFMAX(p->max_diff, fabs(d - p->last));
204 p->diff1_sum += fabs(d - p->last);
205 p->diff1_sum_x2 += (d - p->last) * (d - p->last);
210 if (p->nb_samples >= s->tc_samples) {
211 p->max_sigma_x2 = FFMAX(p->max_sigma_x2, p->avg_sigma_x2);
212 p->min_sigma_x2 = FFMIN(p->min_sigma_x2, p->avg_sigma_x2);
217 static void set_meta(AVDictionary **metadata, int chan, const char *key,
218 const char *fmt, double val)
223 snprintf(value, sizeof(value), fmt, val);
225 snprintf(key2, sizeof(key2), "lavfi.astats.%d.%s", chan, key);
227 snprintf(key2, sizeof(key2), "lavfi.astats.%s", key);
228 av_dict_set(metadata, key2, value, 0);
231 #define LINEAR_TO_DB(x) (log10(x) * 20)
233 static void set_metadata(AudioStatsContext *s, AVDictionary **metadata)
235 uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0;
236 double min_runs = 0, max_runs = 0,
237 min = DBL_MAX, max = DBL_MIN, min_diff = DBL_MAX, max_diff = 0,
238 nmin = DBL_MAX, nmax = DBL_MIN,
244 min_sigma_x2 = DBL_MAX,
245 max_sigma_x2 = DBL_MIN;
249 for (c = 0; c < s->nb_channels; c++) {
250 ChannelStats *p = &s->chstats[c];
252 if (p->nb_samples < s->tc_samples)
253 p->min_sigma_x2 = p->max_sigma_x2 = p->sigma_x2 / p->nb_samples;
255 min = FFMIN(min, p->min);
256 max = FFMAX(max, p->max);
257 nmin = FFMIN(nmin, p->nmin);
258 nmax = FFMAX(nmax, p->nmax);
259 min_diff = FFMIN(min_diff, p->min_diff);
260 max_diff = FFMAX(max_diff, p->max_diff);
261 diff1_sum += p->diff1_sum;
262 diff1_sum_x2 += p->diff1_sum_x2;
263 min_sigma_x2 = FFMIN(min_sigma_x2, p->min_sigma_x2);
264 max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
265 sigma_x += p->sigma_x;
266 sigma_x2 += p->sigma_x2;
267 min_count += p->min_count;
268 max_count += p->max_count;
269 min_runs += p->min_runs;
270 max_runs += p->max_runs;
273 nb_samples += p->nb_samples;
274 if (fabs(p->sigma_x) > fabs(max_sigma_x))
275 max_sigma_x = p->sigma_x;
277 set_meta(metadata, c + 1, "DC_offset", "%f", p->sigma_x / p->nb_samples);
278 set_meta(metadata, c + 1, "Min_level", "%f", p->min);
279 set_meta(metadata, c + 1, "Max_level", "%f", p->max);
280 set_meta(metadata, c + 1, "Min_difference", "%f", p->min_diff);
281 set_meta(metadata, c + 1, "Max_difference", "%f", p->max_diff);
282 set_meta(metadata, c + 1, "Mean_difference", "%f", p->diff1_sum / (p->nb_samples - 1));
283 set_meta(metadata, c + 1, "RMS_difference", "%f", sqrt(p->diff1_sum_x2 / (p->nb_samples - 1)));
284 set_meta(metadata, c + 1, "Peak_level", "%f", LINEAR_TO_DB(FFMAX(-p->nmin, p->nmax)));
285 set_meta(metadata, c + 1, "RMS_level", "%f", LINEAR_TO_DB(sqrt(p->sigma_x2 / p->nb_samples)));
286 set_meta(metadata, c + 1, "RMS_peak", "%f", LINEAR_TO_DB(sqrt(p->max_sigma_x2)));
287 set_meta(metadata, c + 1, "RMS_trough", "%f", LINEAR_TO_DB(sqrt(p->min_sigma_x2)));
288 set_meta(metadata, c + 1, "Crest_factor", "%f", p->sigma_x2 ? FFMAX(-p->min, p->max) / sqrt(p->sigma_x2 / p->nb_samples) : 1);
289 set_meta(metadata, c + 1, "Flat_factor", "%f", LINEAR_TO_DB((p->min_runs + p->max_runs) / (p->min_count + p->max_count)));
290 set_meta(metadata, c + 1, "Peak_count", "%f", (float)(p->min_count + p->max_count));
291 bit_depth(s, p->mask, p->imask, &depth);
292 set_meta(metadata, c + 1, "Bit_depth", "%f", depth.num);
293 set_meta(metadata, c + 1, "Bit_depth2", "%f", depth.den);
294 set_meta(metadata, c + 1, "Dynamic_range", "%f", LINEAR_TO_DB(2 * FFMAX(FFABS(p->min), FFABS(p->max))/ p->min_non_zero));
297 set_meta(metadata, 0, "Overall.DC_offset", "%f", max_sigma_x / (nb_samples / s->nb_channels));
298 set_meta(metadata, 0, "Overall.Min_level", "%f", min);
299 set_meta(metadata, 0, "Overall.Max_level", "%f", max);
300 set_meta(metadata, 0, "Overall.Min_difference", "%f", min_diff);
301 set_meta(metadata, 0, "Overall.Max_difference", "%f", max_diff);
302 set_meta(metadata, 0, "Overall.Mean_difference", "%f", diff1_sum / (nb_samples - s->nb_channels));
303 set_meta(metadata, 0, "Overall.RMS_difference", "%f", sqrt(diff1_sum_x2 / (nb_samples - s->nb_channels)));
304 set_meta(metadata, 0, "Overall.Peak_level", "%f", LINEAR_TO_DB(FFMAX(-nmin, nmax)));
305 set_meta(metadata, 0, "Overall.RMS_level", "%f", LINEAR_TO_DB(sqrt(sigma_x2 / nb_samples)));
306 set_meta(metadata, 0, "Overall.RMS_peak", "%f", LINEAR_TO_DB(sqrt(max_sigma_x2)));
307 set_meta(metadata, 0, "Overall.RMS_trough", "%f", LINEAR_TO_DB(sqrt(min_sigma_x2)));
308 set_meta(metadata, 0, "Overall.Flat_factor", "%f", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
309 set_meta(metadata, 0, "Overall.Peak_count", "%f", (float)(min_count + max_count) / (double)s->nb_channels);
310 bit_depth(s, mask, imask, &depth);
311 set_meta(metadata, 0, "Overall.Bit_depth", "%f", depth.num);
312 set_meta(metadata, 0, "Overall.Bit_depth2", "%f", depth.den);
313 set_meta(metadata, 0, "Overall.Number_of_samples", "%f", nb_samples / s->nb_channels);
316 static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
318 AudioStatsContext *s = inlink->dst->priv;
319 AVDictionary **metadata = &buf->metadata;
320 const int channels = s->nb_channels;
323 if (s->reset_count > 0) {
324 if (s->nb_frames >= s->reset_count) {
331 switch (inlink->format) {
332 case AV_SAMPLE_FMT_DBLP:
333 for (c = 0; c < channels; c++) {
334 ChannelStats *p = &s->chstats[c];
335 const double *src = (const double *)buf->extended_data[c];
337 for (i = 0; i < buf->nb_samples; i++, src++)
338 update_stat(s, p, *src, *src, llrint(*src * (UINT64_C(1) << 63)));
341 case AV_SAMPLE_FMT_DBL: {
342 const double *src = (const double *)buf->extended_data[0];
344 for (i = 0; i < buf->nb_samples; i++) {
345 for (c = 0; c < channels; c++, src++)
346 update_stat(s, &s->chstats[c], *src, *src, llrint(*src * (UINT64_C(1) << 63)));
349 case AV_SAMPLE_FMT_FLTP:
350 for (c = 0; c < channels; c++) {
351 ChannelStats *p = &s->chstats[c];
352 const float *src = (const float *)buf->extended_data[c];
354 for (i = 0; i < buf->nb_samples; i++, src++)
355 update_stat(s, p, *src, *src, llrint(*src * (UINT64_C(1) << 31)));
358 case AV_SAMPLE_FMT_FLT: {
359 const float *src = (const float *)buf->extended_data[0];
361 for (i = 0; i < buf->nb_samples; i++) {
362 for (c = 0; c < channels; c++, src++)
363 update_stat(s, &s->chstats[c], *src, *src, llrint(*src * (UINT64_C(1) << 31)));
366 case AV_SAMPLE_FMT_S64P:
367 for (c = 0; c < channels; c++) {
368 ChannelStats *p = &s->chstats[c];
369 const int64_t *src = (const int64_t *)buf->extended_data[c];
371 for (i = 0; i < buf->nb_samples; i++, src++)
372 update_stat(s, p, *src, *src / (double)INT64_MAX, *src);
375 case AV_SAMPLE_FMT_S64: {
376 const int64_t *src = (const int64_t *)buf->extended_data[0];
378 for (i = 0; i < buf->nb_samples; i++) {
379 for (c = 0; c < channels; c++, src++)
380 update_stat(s, &s->chstats[c], *src, *src / (double)INT64_MAX, *src);
383 case AV_SAMPLE_FMT_S32P:
384 for (c = 0; c < channels; c++) {
385 ChannelStats *p = &s->chstats[c];
386 const int32_t *src = (const int32_t *)buf->extended_data[c];
388 for (i = 0; i < buf->nb_samples; i++, src++)
389 update_stat(s, p, *src, *src / (double)INT32_MAX, *src);
392 case AV_SAMPLE_FMT_S32: {
393 const int32_t *src = (const int32_t *)buf->extended_data[0];
395 for (i = 0; i < buf->nb_samples; i++) {
396 for (c = 0; c < channels; c++, src++)
397 update_stat(s, &s->chstats[c], *src, *src / (double)INT32_MAX, *src);
400 case AV_SAMPLE_FMT_S16P:
401 for (c = 0; c < channels; c++) {
402 ChannelStats *p = &s->chstats[c];
403 const int16_t *src = (const int16_t *)buf->extended_data[c];
405 for (i = 0; i < buf->nb_samples; i++, src++)
406 update_stat(s, p, *src, *src / (double)INT16_MAX, *src);
409 case AV_SAMPLE_FMT_S16: {
410 const int16_t *src = (const int16_t *)buf->extended_data[0];
412 for (i = 0; i < buf->nb_samples; i++) {
413 for (c = 0; c < channels; c++, src++)
414 update_stat(s, &s->chstats[c], *src, *src / (double)INT16_MAX, *src);
420 set_metadata(s, metadata);
422 return ff_filter_frame(inlink->dst->outputs[0], buf);
425 static void print_stats(AVFilterContext *ctx)
427 AudioStatsContext *s = ctx->priv;
428 uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0;
429 double min_runs = 0, max_runs = 0,
430 min = DBL_MAX, max = DBL_MIN, min_diff = DBL_MAX, max_diff = 0,
431 nmin = DBL_MAX, nmax = DBL_MIN,
437 min_sigma_x2 = DBL_MAX,
438 max_sigma_x2 = DBL_MIN;
442 for (c = 0; c < s->nb_channels; c++) {
443 ChannelStats *p = &s->chstats[c];
445 if (p->nb_samples < s->tc_samples)
446 p->min_sigma_x2 = p->max_sigma_x2 = p->sigma_x2 / p->nb_samples;
448 min = FFMIN(min, p->min);
449 max = FFMAX(max, p->max);
450 nmin = FFMIN(nmin, p->nmin);
451 nmax = FFMAX(nmax, p->nmax);
452 min_diff = FFMIN(min_diff, p->min_diff);
453 max_diff = FFMAX(max_diff, p->max_diff);
454 diff1_sum_x2 += p->diff1_sum_x2;
455 diff1_sum += p->diff1_sum;
456 min_sigma_x2 = FFMIN(min_sigma_x2, p->min_sigma_x2);
457 max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
458 sigma_x += p->sigma_x;
459 sigma_x2 += p->sigma_x2;
460 min_count += p->min_count;
461 max_count += p->max_count;
462 min_runs += p->min_runs;
463 max_runs += p->max_runs;
466 nb_samples += p->nb_samples;
467 if (fabs(p->sigma_x) > fabs(max_sigma_x))
468 max_sigma_x = p->sigma_x;
470 av_log(ctx, AV_LOG_INFO, "Channel: %d\n", c + 1);
471 av_log(ctx, AV_LOG_INFO, "DC offset: %f\n", p->sigma_x / p->nb_samples);
472 av_log(ctx, AV_LOG_INFO, "Min level: %f\n", p->min);
473 av_log(ctx, AV_LOG_INFO, "Max level: %f\n", p->max);
474 av_log(ctx, AV_LOG_INFO, "Min difference: %f\n", p->min_diff);
475 av_log(ctx, AV_LOG_INFO, "Max difference: %f\n", p->max_diff);
476 av_log(ctx, AV_LOG_INFO, "Mean difference: %f\n", p->diff1_sum / (p->nb_samples - 1));
477 av_log(ctx, AV_LOG_INFO, "RMS difference: %f\n", sqrt(p->diff1_sum_x2 / (p->nb_samples - 1)));
478 av_log(ctx, AV_LOG_INFO, "Peak level dB: %f\n", LINEAR_TO_DB(FFMAX(-p->nmin, p->nmax)));
479 av_log(ctx, AV_LOG_INFO, "RMS level dB: %f\n", LINEAR_TO_DB(sqrt(p->sigma_x2 / p->nb_samples)));
480 av_log(ctx, AV_LOG_INFO, "RMS peak dB: %f\n", LINEAR_TO_DB(sqrt(p->max_sigma_x2)));
481 if (p->min_sigma_x2 != 1)
482 av_log(ctx, AV_LOG_INFO, "RMS trough dB: %f\n",LINEAR_TO_DB(sqrt(p->min_sigma_x2)));
483 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);
484 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)));
485 av_log(ctx, AV_LOG_INFO, "Peak count: %"PRId64"\n", p->min_count + p->max_count);
486 bit_depth(s, p->mask, p->imask, &depth);
487 av_log(ctx, AV_LOG_INFO, "Bit depth: %u/%u\n", depth.num, depth.den);
488 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));
491 av_log(ctx, AV_LOG_INFO, "Overall\n");
492 av_log(ctx, AV_LOG_INFO, "DC offset: %f\n", max_sigma_x / (nb_samples / s->nb_channels));
493 av_log(ctx, AV_LOG_INFO, "Min level: %f\n", min);
494 av_log(ctx, AV_LOG_INFO, "Max level: %f\n", max);
495 av_log(ctx, AV_LOG_INFO, "Min difference: %f\n", min_diff);
496 av_log(ctx, AV_LOG_INFO, "Max difference: %f\n", max_diff);
497 av_log(ctx, AV_LOG_INFO, "Mean difference: %f\n", diff1_sum / (nb_samples - s->nb_channels));
498 av_log(ctx, AV_LOG_INFO, "RMS difference: %f\n", sqrt(diff1_sum_x2 / (nb_samples - s->nb_channels)));
499 av_log(ctx, AV_LOG_INFO, "Peak level dB: %f\n", LINEAR_TO_DB(FFMAX(-nmin, nmax)));
500 av_log(ctx, AV_LOG_INFO, "RMS level dB: %f\n", LINEAR_TO_DB(sqrt(sigma_x2 / nb_samples)));
501 av_log(ctx, AV_LOG_INFO, "RMS peak dB: %f\n", LINEAR_TO_DB(sqrt(max_sigma_x2)));
502 if (min_sigma_x2 != 1)
503 av_log(ctx, AV_LOG_INFO, "RMS trough dB: %f\n", LINEAR_TO_DB(sqrt(min_sigma_x2)));
504 av_log(ctx, AV_LOG_INFO, "Flat factor: %f\n", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
505 av_log(ctx, AV_LOG_INFO, "Peak count: %f\n", (min_count + max_count) / (double)s->nb_channels);
506 bit_depth(s, mask, imask, &depth);
507 av_log(ctx, AV_LOG_INFO, "Bit depth: %u/%u\n", depth.num, depth.den);
508 av_log(ctx, AV_LOG_INFO, "Number of samples: %"PRId64"\n", nb_samples / s->nb_channels);
511 static av_cold void uninit(AVFilterContext *ctx)
513 AudioStatsContext *s = ctx->priv;
517 av_freep(&s->chstats);
520 static const AVFilterPad astats_inputs[] = {
523 .type = AVMEDIA_TYPE_AUDIO,
524 .filter_frame = filter_frame,
529 static const AVFilterPad astats_outputs[] = {
532 .type = AVMEDIA_TYPE_AUDIO,
533 .config_props = config_output,
538 AVFilter ff_af_astats = {
540 .description = NULL_IF_CONFIG_SMALL("Show time domain statistics about audio frames."),
541 .query_formats = query_formats,
542 .priv_size = sizeof(AudioStatsContext),
543 .priv_class = &astats_class,
545 .inputs = astats_inputs,
546 .outputs = astats_outputs,