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
25 #include "libavutil/opt.h"
30 #define MEASURE_ALL UINT_MAX
31 #define MEASURE_NONE 0
33 #define MEASURE_DC_OFFSET (1 << 0)
34 #define MEASURE_MIN_LEVEL (1 << 1)
35 #define MEASURE_MAX_LEVEL (1 << 2)
36 #define MEASURE_MIN_DIFFERENCE (1 << 3)
37 #define MEASURE_MAX_DIFFERENCE (1 << 4)
38 #define MEASURE_MEAN_DIFFERENCE (1 << 5)
39 #define MEASURE_RMS_DIFFERENCE (1 << 6)
40 #define MEASURE_PEAK_LEVEL (1 << 7)
41 #define MEASURE_RMS_LEVEL (1 << 8)
42 #define MEASURE_RMS_PEAK (1 << 9)
43 #define MEASURE_RMS_TROUGH (1 << 10)
44 #define MEASURE_CREST_FACTOR (1 << 11)
45 #define MEASURE_FLAT_FACTOR (1 << 12)
46 #define MEASURE_PEAK_COUNT (1 << 13)
47 #define MEASURE_BIT_DEPTH (1 << 14)
48 #define MEASURE_DYNAMIC_RANGE (1 << 15)
49 #define MEASURE_ZERO_CROSSINGS (1 << 16)
50 #define MEASURE_ZERO_CROSSINGS_RATE (1 << 17)
51 #define MEASURE_NUMBER_OF_SAMPLES (1 << 18)
52 #define MEASURE_NUMBER_OF_NANS (1 << 19)
53 #define MEASURE_NUMBER_OF_INFS (1 << 20)
54 #define MEASURE_NUMBER_OF_DENORMALS (1 << 21)
56 #define MEASURE_MINMAXPEAK (MEASURE_MIN_LEVEL | MEASURE_MAX_LEVEL | MEASURE_PEAK_LEVEL)
58 typedef struct ChannelStats {
62 double sigma_x, sigma_x2;
63 double avg_sigma_x2, min_sigma_x2, max_sigma_x2;
66 double min_run, max_run;
67 double min_runs, max_runs;
68 double min_diff, max_diff;
72 uint64_t min_count, max_count;
77 uint64_t nb_denormals;
80 typedef struct AudioStatsContext {
82 ChannelStats *chstats;
91 int measure_perchannel;
97 #define OFFSET(x) offsetof(AudioStatsContext, x)
98 #define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
100 static const AVOption astats_options[] = {
101 { "length", "set the window length", OFFSET(time_constant), AV_OPT_TYPE_DOUBLE, {.dbl=.05}, .01, 10, FLAGS },
102 { "metadata", "inject metadata in the filtergraph", OFFSET(metadata), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS },
103 { "reset", "recalculate stats after this many frames", OFFSET(reset_count), AV_OPT_TYPE_INT, {.i64=0}, 0, INT_MAX, FLAGS },
104 { "measure_perchannel", "only measure_perchannel these per-channel statistics", OFFSET(measure_perchannel), AV_OPT_TYPE_FLAGS, {.i64=MEASURE_ALL}, 0, UINT_MAX, FLAGS, "measure" },
105 { "none" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_NONE }, 0, 0, FLAGS, "measure" },
106 { "all" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_ALL }, 0, 0, FLAGS, "measure" },
107 { "DC_offset" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_DC_OFFSET }, 0, 0, FLAGS, "measure" },
108 { "Min_level" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_MIN_LEVEL }, 0, 0, FLAGS, "measure" },
109 { "Max_level" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_MAX_LEVEL }, 0, 0, FLAGS, "measure" },
110 { "Min_difference" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_MIN_DIFFERENCE }, 0, 0, FLAGS, "measure" },
111 { "Max_difference" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_MAX_DIFFERENCE }, 0, 0, FLAGS, "measure" },
112 { "Mean_difference" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_MEAN_DIFFERENCE }, 0, 0, FLAGS, "measure" },
113 { "RMS_difference" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_RMS_DIFFERENCE }, 0, 0, FLAGS, "measure" },
114 { "Peak_level" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_PEAK_LEVEL }, 0, 0, FLAGS, "measure" },
115 { "RMS_level" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_RMS_LEVEL }, 0, 0, FLAGS, "measure" },
116 { "RMS_peak" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_RMS_PEAK }, 0, 0, FLAGS, "measure" },
117 { "RMS_trough" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_RMS_TROUGH }, 0, 0, FLAGS, "measure" },
118 { "Crest_factor" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_CREST_FACTOR }, 0, 0, FLAGS, "measure" },
119 { "Flat_factor" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_FLAT_FACTOR }, 0, 0, FLAGS, "measure" },
120 { "Peak_count" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_PEAK_COUNT }, 0, 0, FLAGS, "measure" },
121 { "Bit_depth" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_BIT_DEPTH }, 0, 0, FLAGS, "measure" },
122 { "Dynamic_range" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_DYNAMIC_RANGE }, 0, 0, FLAGS, "measure" },
123 { "Zero_crossings" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_ZERO_CROSSINGS }, 0, 0, FLAGS, "measure" },
124 { "Zero_crossings_rate" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_ZERO_CROSSINGS_RATE }, 0, 0, FLAGS, "measure" },
125 { "Number_of_samples" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_NUMBER_OF_SAMPLES }, 0, 0, FLAGS, "measure" },
126 { "Number_of_NaNs" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_NUMBER_OF_NANS }, 0, 0, FLAGS, "measure" },
127 { "Number_of_Infs" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_NUMBER_OF_INFS }, 0, 0, FLAGS, "measure" },
128 { "Number_of_denormals" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_NUMBER_OF_DENORMALS }, 0, 0, FLAGS, "measure" },
129 { "measure_overall", "only measure_perchannel these overall statistics", OFFSET(measure_overall), AV_OPT_TYPE_FLAGS, {.i64=MEASURE_ALL}, 0, UINT_MAX, FLAGS, "measure" },
133 AVFILTER_DEFINE_CLASS(astats);
135 static int query_formats(AVFilterContext *ctx)
137 AVFilterFormats *formats;
138 AVFilterChannelLayouts *layouts;
139 static const enum AVSampleFormat sample_fmts[] = {
140 AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P,
141 AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_S32P,
142 AV_SAMPLE_FMT_S64, AV_SAMPLE_FMT_S64P,
143 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
144 AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_DBLP,
149 layouts = ff_all_channel_counts();
151 return AVERROR(ENOMEM);
152 ret = ff_set_common_channel_layouts(ctx, layouts);
156 formats = ff_make_format_list(sample_fmts);
158 return AVERROR(ENOMEM);
159 ret = ff_set_common_formats(ctx, formats);
163 formats = ff_all_samplerates();
165 return AVERROR(ENOMEM);
166 return ff_set_common_samplerates(ctx, formats);
169 static void reset_stats(AudioStatsContext *s)
173 for (c = 0; c < s->nb_channels; c++) {
174 ChannelStats *p = &s->chstats[c];
176 p->min = p->nmin = p->min_sigma_x2 = DBL_MAX;
177 p->max = p->nmax = p->max_sigma_x2 = DBL_MIN;
178 p->min_non_zero = DBL_MAX;
179 p->min_diff = DBL_MAX;
180 p->max_diff = DBL_MIN;
191 p->imask = 0xFFFFFFFFFFFFFFFF;
202 static int config_output(AVFilterLink *outlink)
204 AudioStatsContext *s = outlink->src->priv;
206 s->chstats = av_calloc(sizeof(*s->chstats), outlink->channels);
208 return AVERROR(ENOMEM);
209 s->nb_channels = outlink->channels;
210 s->mult = exp((-1 / s->time_constant / outlink->sample_rate));
211 s->tc_samples = 5 * s->time_constant * outlink->sample_rate + .5;
213 s->maxbitdepth = av_get_bytes_per_sample(outlink->format) * 8;
214 s->is_double = outlink->format == AV_SAMPLE_FMT_DBL ||
215 outlink->format == AV_SAMPLE_FMT_DBLP;
217 s->is_float = outlink->format == AV_SAMPLE_FMT_FLT ||
218 outlink->format == AV_SAMPLE_FMT_FLTP;
225 static void bit_depth(AudioStatsContext *s, uint64_t mask, uint64_t imask, AVRational *depth)
227 unsigned result = s->maxbitdepth;
229 mask = mask & (~imask);
231 for (; result && !(mask & 1); --result, mask >>= 1);
236 for (; result; --result, mask >>= 1)
241 static inline void update_minmax(AudioStatsContext *s, ChannelStats *p, double d)
249 static inline void update_stat(AudioStatsContext *s, ChannelStats *p, double d, double nd, int64_t i)
257 } else if (d == p->min) {
259 p->min_run = d == p->last ? p->min_run + 1 : 1;
260 } else if (p->last == p->min) {
261 p->min_runs += p->min_run * p->min_run;
264 if (d != 0 && FFABS(d) < p->min_non_zero)
265 p->min_non_zero = FFABS(d);
273 } else if (d == p->max) {
275 p->max_run = d == p->last ? p->max_run + 1 : 1;
276 } else if (p->last == p->max) {
277 p->max_runs += p->max_run * p->max_run;
281 p->zero_runs += FFSIGN(d) != FFSIGN(p->last_non_zero);
282 p->last_non_zero = d;
286 p->sigma_x2 += nd * nd;
287 p->avg_sigma_x2 = p->avg_sigma_x2 * s->mult + (1.0 - s->mult) * nd * nd;
288 p->min_diff = FFMIN(p->min_diff, fabs(d - p->last));
289 p->max_diff = FFMAX(p->max_diff, fabs(d - p->last));
290 p->diff1_sum += fabs(d - p->last);
291 p->diff1_sum_x2 += (d - p->last) * (d - p->last);
296 if (p->nb_samples >= s->tc_samples) {
297 p->max_sigma_x2 = FFMAX(p->max_sigma_x2, p->avg_sigma_x2);
298 p->min_sigma_x2 = FFMIN(p->min_sigma_x2, p->avg_sigma_x2);
303 static inline void update_float_stat(AudioStatsContext *s, ChannelStats *p, float d)
305 int type = fpclassify(d);
307 p->nb_nans += type == FP_NAN;
308 p->nb_infs += type == FP_INFINITE;
309 p->nb_denormals += type == FP_SUBNORMAL;
312 static inline void update_double_stat(AudioStatsContext *s, ChannelStats *p, double d)
314 int type = fpclassify(d);
316 p->nb_nans += type == FP_NAN;
317 p->nb_infs += type == FP_INFINITE;
318 p->nb_denormals += type == FP_SUBNORMAL;
321 static void set_meta(AVDictionary **metadata, int chan, const char *key,
322 const char *fmt, double val)
327 snprintf(value, sizeof(value), fmt, val);
329 snprintf(key2, sizeof(key2), "lavfi.astats.%d.%s", chan, key);
331 snprintf(key2, sizeof(key2), "lavfi.astats.%s", key);
332 av_dict_set(metadata, key2, value, 0);
335 #define LINEAR_TO_DB(x) (log10(x) * 20)
337 static void set_metadata(AudioStatsContext *s, AVDictionary **metadata)
339 uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0;
340 uint64_t nb_nans = 0, nb_infs = 0, nb_denormals = 0;
341 double min_runs = 0, max_runs = 0,
342 min = DBL_MAX, max = DBL_MIN, min_diff = DBL_MAX, max_diff = 0,
343 nmin = DBL_MAX, nmax = DBL_MIN,
349 min_sigma_x2 = DBL_MAX,
350 max_sigma_x2 = DBL_MIN;
354 for (c = 0; c < s->nb_channels; c++) {
355 ChannelStats *p = &s->chstats[c];
357 if (p->nb_samples < s->tc_samples)
358 p->min_sigma_x2 = p->max_sigma_x2 = p->sigma_x2 / p->nb_samples;
360 min = FFMIN(min, p->min);
361 max = FFMAX(max, p->max);
362 nmin = FFMIN(nmin, p->nmin);
363 nmax = FFMAX(nmax, p->nmax);
364 min_diff = FFMIN(min_diff, p->min_diff);
365 max_diff = FFMAX(max_diff, p->max_diff);
366 diff1_sum += p->diff1_sum;
367 diff1_sum_x2 += p->diff1_sum_x2;
368 min_sigma_x2 = FFMIN(min_sigma_x2, p->min_sigma_x2);
369 max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
370 sigma_x += p->sigma_x;
371 sigma_x2 += p->sigma_x2;
372 min_count += p->min_count;
373 max_count += p->max_count;
374 min_runs += p->min_runs;
375 max_runs += p->max_runs;
378 nb_samples += p->nb_samples;
379 nb_nans += p->nb_nans;
380 nb_infs += p->nb_infs;
381 nb_denormals += p->nb_denormals;
382 if (fabs(p->sigma_x) > fabs(max_sigma_x))
383 max_sigma_x = p->sigma_x;
385 if (s->measure_perchannel & MEASURE_DC_OFFSET)
386 set_meta(metadata, c + 1, "DC_offset", "%f", p->sigma_x / p->nb_samples);
387 if (s->measure_perchannel & MEASURE_MIN_LEVEL)
388 set_meta(metadata, c + 1, "Min_level", "%f", p->min);
389 if (s->measure_perchannel & MEASURE_MAX_LEVEL)
390 set_meta(metadata, c + 1, "Max_level", "%f", p->max);
391 if (s->measure_perchannel & MEASURE_MIN_DIFFERENCE)
392 set_meta(metadata, c + 1, "Min_difference", "%f", p->min_diff);
393 if (s->measure_perchannel & MEASURE_MAX_DIFFERENCE)
394 set_meta(metadata, c + 1, "Max_difference", "%f", p->max_diff);
395 if (s->measure_perchannel & MEASURE_MEAN_DIFFERENCE)
396 set_meta(metadata, c + 1, "Mean_difference", "%f", p->diff1_sum / (p->nb_samples - 1));
397 if (s->measure_perchannel & MEASURE_RMS_DIFFERENCE)
398 set_meta(metadata, c + 1, "RMS_difference", "%f", sqrt(p->diff1_sum_x2 / (p->nb_samples - 1)));
399 if (s->measure_perchannel & MEASURE_PEAK_LEVEL)
400 set_meta(metadata, c + 1, "Peak_level", "%f", LINEAR_TO_DB(FFMAX(-p->nmin, p->nmax)));
401 if (s->measure_perchannel & MEASURE_RMS_LEVEL)
402 set_meta(metadata, c + 1, "RMS_level", "%f", LINEAR_TO_DB(sqrt(p->sigma_x2 / p->nb_samples)));
403 if (s->measure_perchannel & MEASURE_RMS_PEAK)
404 set_meta(metadata, c + 1, "RMS_peak", "%f", LINEAR_TO_DB(sqrt(p->max_sigma_x2)));
405 if (s->measure_perchannel & MEASURE_RMS_TROUGH)
406 set_meta(metadata, c + 1, "RMS_trough", "%f", LINEAR_TO_DB(sqrt(p->min_sigma_x2)));
407 if (s->measure_perchannel & MEASURE_CREST_FACTOR)
408 set_meta(metadata, c + 1, "Crest_factor", "%f", p->sigma_x2 ? FFMAX(-p->min, p->max) / sqrt(p->sigma_x2 / p->nb_samples) : 1);
409 if (s->measure_perchannel & MEASURE_FLAT_FACTOR)
410 set_meta(metadata, c + 1, "Flat_factor", "%f", LINEAR_TO_DB((p->min_runs + p->max_runs) / (p->min_count + p->max_count)));
411 if (s->measure_perchannel & MEASURE_PEAK_COUNT)
412 set_meta(metadata, c + 1, "Peak_count", "%f", (float)(p->min_count + p->max_count));
413 if (s->measure_perchannel & MEASURE_BIT_DEPTH) {
414 bit_depth(s, p->mask, p->imask, &depth);
415 set_meta(metadata, c + 1, "Bit_depth", "%f", depth.num);
416 set_meta(metadata, c + 1, "Bit_depth2", "%f", depth.den);
418 if (s->measure_perchannel & MEASURE_DYNAMIC_RANGE)
419 set_meta(metadata, c + 1, "Dynamic_range", "%f", LINEAR_TO_DB(2 * FFMAX(FFABS(p->min), FFABS(p->max))/ p->min_non_zero));
420 if (s->measure_perchannel & MEASURE_ZERO_CROSSINGS)
421 set_meta(metadata, c + 1, "Zero_crossings", "%f", p->zero_runs);
422 if (s->measure_perchannel & MEASURE_ZERO_CROSSINGS_RATE)
423 set_meta(metadata, c + 1, "Zero_crossings_rate", "%f", p->zero_runs/(double)p->nb_samples);
424 if ((s->is_float || s->is_double) && s->measure_perchannel & MEASURE_NUMBER_OF_NANS)
425 set_meta(metadata, c + 1, "Number of NaNs", "%f", p->nb_nans);
426 if ((s->is_float || s->is_double) && s->measure_perchannel & MEASURE_NUMBER_OF_INFS)
427 set_meta(metadata, c + 1, "Number of Infs", "%f", p->nb_infs);
428 if ((s->is_float || s->is_double) && s->measure_perchannel & MEASURE_NUMBER_OF_DENORMALS)
429 set_meta(metadata, c + 1, "Number of denormals", "%f", p->nb_denormals);
432 if (s->measure_overall & MEASURE_DC_OFFSET)
433 set_meta(metadata, 0, "Overall.DC_offset", "%f", max_sigma_x / (nb_samples / s->nb_channels));
434 if (s->measure_overall & MEASURE_MIN_LEVEL)
435 set_meta(metadata, 0, "Overall.Min_level", "%f", min);
436 if (s->measure_overall & MEASURE_MAX_LEVEL)
437 set_meta(metadata, 0, "Overall.Max_level", "%f", max);
438 if (s->measure_overall & MEASURE_MIN_DIFFERENCE)
439 set_meta(metadata, 0, "Overall.Min_difference", "%f", min_diff);
440 if (s->measure_overall & MEASURE_MAX_DIFFERENCE)
441 set_meta(metadata, 0, "Overall.Max_difference", "%f", max_diff);
442 if (s->measure_overall & MEASURE_MEAN_DIFFERENCE)
443 set_meta(metadata, 0, "Overall.Mean_difference", "%f", diff1_sum / (nb_samples - s->nb_channels));
444 if (s->measure_overall & MEASURE_RMS_DIFFERENCE)
445 set_meta(metadata, 0, "Overall.RMS_difference", "%f", sqrt(diff1_sum_x2 / (nb_samples - s->nb_channels)));
446 if (s->measure_overall & MEASURE_PEAK_LEVEL)
447 set_meta(metadata, 0, "Overall.Peak_level", "%f", LINEAR_TO_DB(FFMAX(-nmin, nmax)));
448 if (s->measure_overall & MEASURE_RMS_LEVEL)
449 set_meta(metadata, 0, "Overall.RMS_level", "%f", LINEAR_TO_DB(sqrt(sigma_x2 / nb_samples)));
450 if (s->measure_overall & MEASURE_RMS_PEAK)
451 set_meta(metadata, 0, "Overall.RMS_peak", "%f", LINEAR_TO_DB(sqrt(max_sigma_x2)));
452 if (s->measure_overall & MEASURE_RMS_TROUGH)
453 set_meta(metadata, 0, "Overall.RMS_trough", "%f", LINEAR_TO_DB(sqrt(min_sigma_x2)));
454 if (s->measure_overall & MEASURE_FLAT_FACTOR)
455 set_meta(metadata, 0, "Overall.Flat_factor", "%f", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
456 if (s->measure_overall & MEASURE_PEAK_COUNT)
457 set_meta(metadata, 0, "Overall.Peak_count", "%f", (float)(min_count + max_count) / (double)s->nb_channels);
458 if (s->measure_overall & MEASURE_BIT_DEPTH) {
459 bit_depth(s, mask, imask, &depth);
460 set_meta(metadata, 0, "Overall.Bit_depth", "%f", depth.num);
461 set_meta(metadata, 0, "Overall.Bit_depth2", "%f", depth.den);
463 if (s->measure_overall & MEASURE_NUMBER_OF_SAMPLES)
464 set_meta(metadata, 0, "Overall.Number_of_samples", "%f", nb_samples / s->nb_channels);
465 if ((s->is_float || s->is_double) && s->measure_overall & MEASURE_NUMBER_OF_NANS)
466 set_meta(metadata, 0, "Number of NaNs", "%f", nb_nans / (float)s->nb_channels);
467 if ((s->is_float || s->is_double) && s->measure_overall & MEASURE_NUMBER_OF_INFS)
468 set_meta(metadata, 0, "Number of Infs", "%f", nb_infs / (float)s->nb_channels);
469 if ((s->is_float || s->is_double) && s->measure_overall & MEASURE_NUMBER_OF_DENORMALS)
470 set_meta(metadata, 0, "Number of denormals", "%f", nb_denormals / (float)s->nb_channels);
473 #define UPDATE_STATS_P(type, update_func, update_float, channel_func) \
474 for (int c = 0; c < channels; c++) { \
475 ChannelStats *p = &s->chstats[c]; \
476 const type *src = (const type *)data[c]; \
477 const type * const srcend = src + samples; \
478 for (; src < srcend; src++) { \
485 #define UPDATE_STATS_I(type, update_func, update_float, channel_func) \
486 for (int c = 0; c < channels; c++) { \
487 ChannelStats *p = &s->chstats[c]; \
488 const type *src = (const type *)data[0]; \
489 const type * const srcend = src + samples * channels; \
490 for (src += c; src < srcend; src += channels) { \
497 #define UPDATE_STATS(planar, type, sample, normalizer_suffix, int_sample) \
498 if ((s->measure_overall | s->measure_perchannel) & ~MEASURE_MINMAXPEAK) { \
499 UPDATE_STATS_##planar(type, update_stat(s, p, sample, sample normalizer_suffix, int_sample), s->is_float ? update_float_stat(s, p, sample) : s->is_double ? update_double_stat(s, p, sample) : (void)NULL, ); \
501 UPDATE_STATS_##planar(type, update_minmax(s, p, sample), , p->nmin = p->min normalizer_suffix; p->nmax = p->max normalizer_suffix;); \
504 static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
506 AudioStatsContext *s = inlink->dst->priv;
507 AVDictionary **metadata = &buf->metadata;
508 const int channels = s->nb_channels;
509 const int samples = buf->nb_samples;
510 const uint8_t * const * const data = (const uint8_t * const *)buf->extended_data;
512 if (s->reset_count > 0) {
513 if (s->nb_frames >= s->reset_count) {
520 switch (inlink->format) {
521 case AV_SAMPLE_FMT_DBLP:
522 UPDATE_STATS(P, double, *src, , llrint(*src * (UINT64_C(1) << 63)));
524 case AV_SAMPLE_FMT_DBL:
525 UPDATE_STATS(I, double, *src, , llrint(*src * (UINT64_C(1) << 63)));
527 case AV_SAMPLE_FMT_FLTP:
528 UPDATE_STATS(P, float, *src, , llrint(*src * (UINT64_C(1) << 31)));
530 case AV_SAMPLE_FMT_FLT:
531 UPDATE_STATS(I, float, *src, , llrint(*src * (UINT64_C(1) << 31)));
533 case AV_SAMPLE_FMT_S64P:
534 UPDATE_STATS(P, int64_t, *src, / (double)INT64_MAX, *src);
536 case AV_SAMPLE_FMT_S64:
537 UPDATE_STATS(I, int64_t, *src, / (double)INT64_MAX, *src);
539 case AV_SAMPLE_FMT_S32P:
540 UPDATE_STATS(P, int32_t, *src, / (double)INT32_MAX, *src);
542 case AV_SAMPLE_FMT_S32:
543 UPDATE_STATS(I, int32_t, *src, / (double)INT32_MAX, *src);
545 case AV_SAMPLE_FMT_S16P:
546 UPDATE_STATS(P, int16_t, *src, / (double)INT16_MAX, *src);
548 case AV_SAMPLE_FMT_S16:
549 UPDATE_STATS(I, int16_t, *src, / (double)INT16_MAX, *src);
554 set_metadata(s, metadata);
556 return ff_filter_frame(inlink->dst->outputs[0], buf);
559 static void print_stats(AVFilterContext *ctx)
561 AudioStatsContext *s = ctx->priv;
562 uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0;
563 uint64_t nb_nans = 0, nb_infs = 0, nb_denormals = 0;
564 double min_runs = 0, max_runs = 0,
565 min = DBL_MAX, max = DBL_MIN, min_diff = DBL_MAX, max_diff = 0,
566 nmin = DBL_MAX, nmax = DBL_MIN,
572 min_sigma_x2 = DBL_MAX,
573 max_sigma_x2 = DBL_MIN;
577 for (c = 0; c < s->nb_channels; c++) {
578 ChannelStats *p = &s->chstats[c];
580 if (p->nb_samples < s->tc_samples)
581 p->min_sigma_x2 = p->max_sigma_x2 = p->sigma_x2 / p->nb_samples;
583 min = FFMIN(min, p->min);
584 max = FFMAX(max, p->max);
585 nmin = FFMIN(nmin, p->nmin);
586 nmax = FFMAX(nmax, p->nmax);
587 min_diff = FFMIN(min_diff, p->min_diff);
588 max_diff = FFMAX(max_diff, p->max_diff);
589 diff1_sum_x2 += p->diff1_sum_x2;
590 diff1_sum += p->diff1_sum;
591 min_sigma_x2 = FFMIN(min_sigma_x2, p->min_sigma_x2);
592 max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
593 sigma_x += p->sigma_x;
594 sigma_x2 += p->sigma_x2;
595 min_count += p->min_count;
596 max_count += p->max_count;
597 min_runs += p->min_runs;
598 max_runs += p->max_runs;
601 nb_samples += p->nb_samples;
602 nb_nans += p->nb_nans;
603 nb_infs += p->nb_infs;
604 nb_denormals += p->nb_denormals;
605 if (fabs(p->sigma_x) > fabs(max_sigma_x))
606 max_sigma_x = p->sigma_x;
608 av_log(ctx, AV_LOG_INFO, "Channel: %d\n", c + 1);
609 if (s->measure_perchannel & MEASURE_DC_OFFSET)
610 av_log(ctx, AV_LOG_INFO, "DC offset: %f\n", p->sigma_x / p->nb_samples);
611 if (s->measure_perchannel & MEASURE_MIN_LEVEL)
612 av_log(ctx, AV_LOG_INFO, "Min level: %f\n", p->min);
613 if (s->measure_perchannel & MEASURE_MAX_LEVEL)
614 av_log(ctx, AV_LOG_INFO, "Max level: %f\n", p->max);
615 if (s->measure_perchannel & MEASURE_MIN_DIFFERENCE)
616 av_log(ctx, AV_LOG_INFO, "Min difference: %f\n", p->min_diff);
617 if (s->measure_perchannel & MEASURE_MAX_DIFFERENCE)
618 av_log(ctx, AV_LOG_INFO, "Max difference: %f\n", p->max_diff);
619 if (s->measure_perchannel & MEASURE_MEAN_DIFFERENCE)
620 av_log(ctx, AV_LOG_INFO, "Mean difference: %f\n", p->diff1_sum / (p->nb_samples - 1));
621 if (s->measure_perchannel & MEASURE_RMS_DIFFERENCE)
622 av_log(ctx, AV_LOG_INFO, "RMS difference: %f\n", sqrt(p->diff1_sum_x2 / (p->nb_samples - 1)));
623 if (s->measure_perchannel & MEASURE_PEAK_LEVEL)
624 av_log(ctx, AV_LOG_INFO, "Peak level dB: %f\n", LINEAR_TO_DB(FFMAX(-p->nmin, p->nmax)));
625 if (s->measure_perchannel & MEASURE_RMS_LEVEL)
626 av_log(ctx, AV_LOG_INFO, "RMS level dB: %f\n", LINEAR_TO_DB(sqrt(p->sigma_x2 / p->nb_samples)));
627 if (s->measure_perchannel & MEASURE_RMS_PEAK)
628 av_log(ctx, AV_LOG_INFO, "RMS peak dB: %f\n", LINEAR_TO_DB(sqrt(p->max_sigma_x2)));
629 if (s->measure_perchannel & MEASURE_RMS_TROUGH)
630 if (p->min_sigma_x2 != 1)
631 av_log(ctx, AV_LOG_INFO, "RMS trough dB: %f\n",LINEAR_TO_DB(sqrt(p->min_sigma_x2)));
632 if (s->measure_perchannel & MEASURE_CREST_FACTOR)
633 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);
634 if (s->measure_perchannel & MEASURE_FLAT_FACTOR)
635 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)));
636 if (s->measure_perchannel & MEASURE_PEAK_COUNT)
637 av_log(ctx, AV_LOG_INFO, "Peak count: %"PRId64"\n", p->min_count + p->max_count);
638 if (s->measure_perchannel & MEASURE_BIT_DEPTH) {
639 bit_depth(s, p->mask, p->imask, &depth);
640 av_log(ctx, AV_LOG_INFO, "Bit depth: %u/%u\n", depth.num, depth.den);
642 if (s->measure_perchannel & MEASURE_DYNAMIC_RANGE)
643 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));
644 if (s->measure_perchannel & MEASURE_ZERO_CROSSINGS)
645 av_log(ctx, AV_LOG_INFO, "Zero crossings: %"PRId64"\n", p->zero_runs);
646 if (s->measure_perchannel & MEASURE_ZERO_CROSSINGS_RATE)
647 av_log(ctx, AV_LOG_INFO, "Zero crossings rate: %f\n", p->zero_runs/(double)p->nb_samples);
648 if ((s->is_float || s->is_double) && s->measure_perchannel & MEASURE_NUMBER_OF_NANS)
649 av_log(ctx, AV_LOG_INFO, "Number of NaNs: %"PRId64"\n", p->nb_nans);
650 if ((s->is_float || s->is_double) && s->measure_perchannel & MEASURE_NUMBER_OF_INFS)
651 av_log(ctx, AV_LOG_INFO, "Number of Infs: %"PRId64"\n", p->nb_infs);
652 if ((s->is_float || s->is_double) && s->measure_perchannel & MEASURE_NUMBER_OF_DENORMALS)
653 av_log(ctx, AV_LOG_INFO, "Number of denormals: %"PRId64"\n", p->nb_denormals);
656 av_log(ctx, AV_LOG_INFO, "Overall\n");
657 if (s->measure_overall & MEASURE_DC_OFFSET)
658 av_log(ctx, AV_LOG_INFO, "DC offset: %f\n", max_sigma_x / (nb_samples / s->nb_channels));
659 if (s->measure_overall & MEASURE_MIN_LEVEL)
660 av_log(ctx, AV_LOG_INFO, "Min level: %f\n", min);
661 if (s->measure_overall & MEASURE_MAX_LEVEL)
662 av_log(ctx, AV_LOG_INFO, "Max level: %f\n", max);
663 if (s->measure_overall & MEASURE_MIN_DIFFERENCE)
664 av_log(ctx, AV_LOG_INFO, "Min difference: %f\n", min_diff);
665 if (s->measure_overall & MEASURE_MAX_DIFFERENCE)
666 av_log(ctx, AV_LOG_INFO, "Max difference: %f\n", max_diff);
667 if (s->measure_overall & MEASURE_MEAN_DIFFERENCE)
668 av_log(ctx, AV_LOG_INFO, "Mean difference: %f\n", diff1_sum / (nb_samples - s->nb_channels));
669 if (s->measure_overall & MEASURE_RMS_DIFFERENCE)
670 av_log(ctx, AV_LOG_INFO, "RMS difference: %f\n", sqrt(diff1_sum_x2 / (nb_samples - s->nb_channels)));
671 if (s->measure_overall & MEASURE_PEAK_LEVEL)
672 av_log(ctx, AV_LOG_INFO, "Peak level dB: %f\n", LINEAR_TO_DB(FFMAX(-nmin, nmax)));
673 if (s->measure_overall & MEASURE_RMS_LEVEL)
674 av_log(ctx, AV_LOG_INFO, "RMS level dB: %f\n", LINEAR_TO_DB(sqrt(sigma_x2 / nb_samples)));
675 if (s->measure_overall & MEASURE_RMS_PEAK)
676 av_log(ctx, AV_LOG_INFO, "RMS peak dB: %f\n", LINEAR_TO_DB(sqrt(max_sigma_x2)));
677 if (s->measure_overall & MEASURE_RMS_TROUGH)
678 if (min_sigma_x2 != 1)
679 av_log(ctx, AV_LOG_INFO, "RMS trough dB: %f\n", LINEAR_TO_DB(sqrt(min_sigma_x2)));
680 if (s->measure_overall & MEASURE_FLAT_FACTOR)
681 av_log(ctx, AV_LOG_INFO, "Flat factor: %f\n", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
682 if (s->measure_overall & MEASURE_PEAK_COUNT)
683 av_log(ctx, AV_LOG_INFO, "Peak count: %f\n", (min_count + max_count) / (double)s->nb_channels);
684 if (s->measure_overall & MEASURE_BIT_DEPTH) {
685 bit_depth(s, mask, imask, &depth);
686 av_log(ctx, AV_LOG_INFO, "Bit depth: %u/%u\n", depth.num, depth.den);
688 if (s->measure_overall & MEASURE_NUMBER_OF_SAMPLES)
689 av_log(ctx, AV_LOG_INFO, "Number of samples: %"PRId64"\n", nb_samples / s->nb_channels);
690 if ((s->is_float || s->is_double) && s->measure_overall & MEASURE_NUMBER_OF_NANS)
691 av_log(ctx, AV_LOG_INFO, "Number of NaNs: %f\n", nb_nans / (float)s->nb_channels);
692 if ((s->is_float || s->is_double) && s->measure_overall & MEASURE_NUMBER_OF_INFS)
693 av_log(ctx, AV_LOG_INFO, "Number of Infs: %f\n", nb_infs / (float)s->nb_channels);
694 if ((s->is_float || s->is_double) && s->measure_overall & MEASURE_NUMBER_OF_DENORMALS)
695 av_log(ctx, AV_LOG_INFO, "Number of denormals: %f\n", nb_denormals / (float)s->nb_channels);
698 static av_cold void uninit(AVFilterContext *ctx)
700 AudioStatsContext *s = ctx->priv;
704 av_freep(&s->chstats);
707 static const AVFilterPad astats_inputs[] = {
710 .type = AVMEDIA_TYPE_AUDIO,
711 .filter_frame = filter_frame,
716 static const AVFilterPad astats_outputs[] = {
719 .type = AVMEDIA_TYPE_AUDIO,
720 .config_props = config_output,
725 AVFilter ff_af_astats = {
727 .description = NULL_IF_CONFIG_SMALL("Show time domain statistics about audio frames."),
728 .query_formats = query_formats,
729 .priv_size = sizeof(AudioStatsContext),
730 .priv_class = &astats_class,
732 .inputs = astats_inputs,
733 .outputs = astats_outputs,