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_MAX;
178 p->min_non_zero = DBL_MAX;
179 p->min_diff = DBL_MAX;
191 p->imask = 0xFFFFFFFFFFFFFFFF;
203 static int config_output(AVFilterLink *outlink)
205 AudioStatsContext *s = outlink->src->priv;
207 s->chstats = av_calloc(sizeof(*s->chstats), outlink->channels);
209 return AVERROR(ENOMEM);
210 s->nb_channels = outlink->channels;
211 s->mult = exp((-1 / s->time_constant / outlink->sample_rate));
212 s->tc_samples = 5 * s->time_constant * outlink->sample_rate + .5;
214 s->maxbitdepth = av_get_bytes_per_sample(outlink->format) * 8;
215 s->is_double = outlink->format == AV_SAMPLE_FMT_DBL ||
216 outlink->format == AV_SAMPLE_FMT_DBLP;
218 s->is_float = outlink->format == AV_SAMPLE_FMT_FLT ||
219 outlink->format == AV_SAMPLE_FMT_FLTP;
226 static void bit_depth(AudioStatsContext *s, uint64_t mask, uint64_t imask, AVRational *depth)
228 unsigned result = s->maxbitdepth;
230 mask = mask & (~imask);
232 for (; result && !(mask & 1); --result, mask >>= 1);
237 for (; result; --result, mask >>= 1)
242 static inline void update_minmax(AudioStatsContext *s, ChannelStats *p, double d)
250 static inline void update_stat(AudioStatsContext *s, ChannelStats *p, double d, double nd, int64_t i)
258 } else if (d == p->min) {
260 p->min_run = d == p->last ? p->min_run + 1 : 1;
261 } else if (p->last == p->min) {
262 p->min_runs += p->min_run * p->min_run;
265 if (d != 0 && FFABS(d) < p->min_non_zero)
266 p->min_non_zero = FFABS(d);
274 } else if (d == p->max) {
276 p->max_run = d == p->last ? p->max_run + 1 : 1;
277 } else if (p->last == p->max) {
278 p->max_runs += p->max_run * p->max_run;
282 p->zero_runs += FFSIGN(d) != FFSIGN(p->last_non_zero);
283 p->last_non_zero = d;
287 p->sigma_x2 += nd * nd;
288 p->avg_sigma_x2 = p->avg_sigma_x2 * s->mult + (1.0 - s->mult) * nd * nd;
289 if (!isnan(p->last)) {
290 p->min_diff = FFMIN(p->min_diff, fabs(d - p->last));
291 p->max_diff = FFMAX(p->max_diff, fabs(d - p->last));
292 p->diff1_sum += fabs(d - p->last);
293 p->diff1_sum_x2 += (d - p->last) * (d - p->last);
299 if (p->nb_samples >= s->tc_samples) {
300 p->max_sigma_x2 = FFMAX(p->max_sigma_x2, p->avg_sigma_x2);
301 p->min_sigma_x2 = FFMIN(p->min_sigma_x2, p->avg_sigma_x2);
306 static inline void update_float_stat(AudioStatsContext *s, ChannelStats *p, float d)
308 int type = fpclassify(d);
310 p->nb_nans += type == FP_NAN;
311 p->nb_infs += type == FP_INFINITE;
312 p->nb_denormals += type == FP_SUBNORMAL;
315 static inline void update_double_stat(AudioStatsContext *s, ChannelStats *p, double d)
317 int type = fpclassify(d);
319 p->nb_nans += type == FP_NAN;
320 p->nb_infs += type == FP_INFINITE;
321 p->nb_denormals += type == FP_SUBNORMAL;
324 static void set_meta(AVDictionary **metadata, int chan, const char *key,
325 const char *fmt, double val)
330 snprintf(value, sizeof(value), fmt, val);
332 snprintf(key2, sizeof(key2), "lavfi.astats.%d.%s", chan, key);
334 snprintf(key2, sizeof(key2), "lavfi.astats.%s", key);
335 av_dict_set(metadata, key2, value, 0);
338 #define LINEAR_TO_DB(x) (log10(x) * 20)
340 static void set_metadata(AudioStatsContext *s, AVDictionary **metadata)
342 uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0;
343 uint64_t nb_nans = 0, nb_infs = 0, nb_denormals = 0;
344 double min_runs = 0, max_runs = 0,
345 min = DBL_MAX, max =-DBL_MAX, min_diff = DBL_MAX, max_diff = 0,
346 nmin = DBL_MAX, nmax =-DBL_MAX,
352 min_sigma_x2 = DBL_MAX,
353 max_sigma_x2 =-DBL_MAX;
357 for (c = 0; c < s->nb_channels; c++) {
358 ChannelStats *p = &s->chstats[c];
360 if (p->nb_samples < s->tc_samples)
361 p->min_sigma_x2 = p->max_sigma_x2 = p->sigma_x2 / p->nb_samples;
363 min = FFMIN(min, p->min);
364 max = FFMAX(max, p->max);
365 nmin = FFMIN(nmin, p->nmin);
366 nmax = FFMAX(nmax, p->nmax);
367 min_diff = FFMIN(min_diff, p->min_diff);
368 max_diff = FFMAX(max_diff, p->max_diff);
369 diff1_sum += p->diff1_sum;
370 diff1_sum_x2 += p->diff1_sum_x2;
371 min_sigma_x2 = FFMIN(min_sigma_x2, p->min_sigma_x2);
372 max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
373 sigma_x += p->sigma_x;
374 sigma_x2 += p->sigma_x2;
375 min_count += p->min_count;
376 max_count += p->max_count;
377 min_runs += p->min_runs;
378 max_runs += p->max_runs;
381 nb_samples += p->nb_samples;
382 nb_nans += p->nb_nans;
383 nb_infs += p->nb_infs;
384 nb_denormals += p->nb_denormals;
385 if (fabs(p->sigma_x) > fabs(max_sigma_x))
386 max_sigma_x = p->sigma_x;
388 if (s->measure_perchannel & MEASURE_DC_OFFSET)
389 set_meta(metadata, c + 1, "DC_offset", "%f", p->sigma_x / p->nb_samples);
390 if (s->measure_perchannel & MEASURE_MIN_LEVEL)
391 set_meta(metadata, c + 1, "Min_level", "%f", p->min);
392 if (s->measure_perchannel & MEASURE_MAX_LEVEL)
393 set_meta(metadata, c + 1, "Max_level", "%f", p->max);
394 if (s->measure_perchannel & MEASURE_MIN_DIFFERENCE)
395 set_meta(metadata, c + 1, "Min_difference", "%f", p->min_diff);
396 if (s->measure_perchannel & MEASURE_MAX_DIFFERENCE)
397 set_meta(metadata, c + 1, "Max_difference", "%f", p->max_diff);
398 if (s->measure_perchannel & MEASURE_MEAN_DIFFERENCE)
399 set_meta(metadata, c + 1, "Mean_difference", "%f", p->diff1_sum / (p->nb_samples - 1));
400 if (s->measure_perchannel & MEASURE_RMS_DIFFERENCE)
401 set_meta(metadata, c + 1, "RMS_difference", "%f", sqrt(p->diff1_sum_x2 / (p->nb_samples - 1)));
402 if (s->measure_perchannel & MEASURE_PEAK_LEVEL)
403 set_meta(metadata, c + 1, "Peak_level", "%f", LINEAR_TO_DB(FFMAX(-p->nmin, p->nmax)));
404 if (s->measure_perchannel & MEASURE_RMS_LEVEL)
405 set_meta(metadata, c + 1, "RMS_level", "%f", LINEAR_TO_DB(sqrt(p->sigma_x2 / p->nb_samples)));
406 if (s->measure_perchannel & MEASURE_RMS_PEAK)
407 set_meta(metadata, c + 1, "RMS_peak", "%f", LINEAR_TO_DB(sqrt(p->max_sigma_x2)));
408 if (s->measure_perchannel & MEASURE_RMS_TROUGH)
409 set_meta(metadata, c + 1, "RMS_trough", "%f", LINEAR_TO_DB(sqrt(p->min_sigma_x2)));
410 if (s->measure_perchannel & MEASURE_CREST_FACTOR)
411 set_meta(metadata, c + 1, "Crest_factor", "%f", p->sigma_x2 ? FFMAX(-p->min, p->max) / sqrt(p->sigma_x2 / p->nb_samples) : 1);
412 if (s->measure_perchannel & MEASURE_FLAT_FACTOR)
413 set_meta(metadata, c + 1, "Flat_factor", "%f", LINEAR_TO_DB((p->min_runs + p->max_runs) / (p->min_count + p->max_count)));
414 if (s->measure_perchannel & MEASURE_PEAK_COUNT)
415 set_meta(metadata, c + 1, "Peak_count", "%f", (float)(p->min_count + p->max_count));
416 if (s->measure_perchannel & MEASURE_BIT_DEPTH) {
417 bit_depth(s, p->mask, p->imask, &depth);
418 set_meta(metadata, c + 1, "Bit_depth", "%f", depth.num);
419 set_meta(metadata, c + 1, "Bit_depth2", "%f", depth.den);
421 if (s->measure_perchannel & MEASURE_DYNAMIC_RANGE)
422 set_meta(metadata, c + 1, "Dynamic_range", "%f", LINEAR_TO_DB(2 * FFMAX(FFABS(p->min), FFABS(p->max))/ p->min_non_zero));
423 if (s->measure_perchannel & MEASURE_ZERO_CROSSINGS)
424 set_meta(metadata, c + 1, "Zero_crossings", "%f", p->zero_runs);
425 if (s->measure_perchannel & MEASURE_ZERO_CROSSINGS_RATE)
426 set_meta(metadata, c + 1, "Zero_crossings_rate", "%f", p->zero_runs/(double)p->nb_samples);
427 if ((s->is_float || s->is_double) && s->measure_perchannel & MEASURE_NUMBER_OF_NANS)
428 set_meta(metadata, c + 1, "Number of NaNs", "%f", p->nb_nans);
429 if ((s->is_float || s->is_double) && s->measure_perchannel & MEASURE_NUMBER_OF_INFS)
430 set_meta(metadata, c + 1, "Number of Infs", "%f", p->nb_infs);
431 if ((s->is_float || s->is_double) && s->measure_perchannel & MEASURE_NUMBER_OF_DENORMALS)
432 set_meta(metadata, c + 1, "Number of denormals", "%f", p->nb_denormals);
435 if (s->measure_overall & MEASURE_DC_OFFSET)
436 set_meta(metadata, 0, "Overall.DC_offset", "%f", max_sigma_x / (nb_samples / s->nb_channels));
437 if (s->measure_overall & MEASURE_MIN_LEVEL)
438 set_meta(metadata, 0, "Overall.Min_level", "%f", min);
439 if (s->measure_overall & MEASURE_MAX_LEVEL)
440 set_meta(metadata, 0, "Overall.Max_level", "%f", max);
441 if (s->measure_overall & MEASURE_MIN_DIFFERENCE)
442 set_meta(metadata, 0, "Overall.Min_difference", "%f", min_diff);
443 if (s->measure_overall & MEASURE_MAX_DIFFERENCE)
444 set_meta(metadata, 0, "Overall.Max_difference", "%f", max_diff);
445 if (s->measure_overall & MEASURE_MEAN_DIFFERENCE)
446 set_meta(metadata, 0, "Overall.Mean_difference", "%f", diff1_sum / (nb_samples - s->nb_channels));
447 if (s->measure_overall & MEASURE_RMS_DIFFERENCE)
448 set_meta(metadata, 0, "Overall.RMS_difference", "%f", sqrt(diff1_sum_x2 / (nb_samples - s->nb_channels)));
449 if (s->measure_overall & MEASURE_PEAK_LEVEL)
450 set_meta(metadata, 0, "Overall.Peak_level", "%f", LINEAR_TO_DB(FFMAX(-nmin, nmax)));
451 if (s->measure_overall & MEASURE_RMS_LEVEL)
452 set_meta(metadata, 0, "Overall.RMS_level", "%f", LINEAR_TO_DB(sqrt(sigma_x2 / nb_samples)));
453 if (s->measure_overall & MEASURE_RMS_PEAK)
454 set_meta(metadata, 0, "Overall.RMS_peak", "%f", LINEAR_TO_DB(sqrt(max_sigma_x2)));
455 if (s->measure_overall & MEASURE_RMS_TROUGH)
456 set_meta(metadata, 0, "Overall.RMS_trough", "%f", LINEAR_TO_DB(sqrt(min_sigma_x2)));
457 if (s->measure_overall & MEASURE_FLAT_FACTOR)
458 set_meta(metadata, 0, "Overall.Flat_factor", "%f", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
459 if (s->measure_overall & MEASURE_PEAK_COUNT)
460 set_meta(metadata, 0, "Overall.Peak_count", "%f", (float)(min_count + max_count) / (double)s->nb_channels);
461 if (s->measure_overall & MEASURE_BIT_DEPTH) {
462 bit_depth(s, mask, imask, &depth);
463 set_meta(metadata, 0, "Overall.Bit_depth", "%f", depth.num);
464 set_meta(metadata, 0, "Overall.Bit_depth2", "%f", depth.den);
466 if (s->measure_overall & MEASURE_NUMBER_OF_SAMPLES)
467 set_meta(metadata, 0, "Overall.Number_of_samples", "%f", nb_samples / s->nb_channels);
468 if ((s->is_float || s->is_double) && s->measure_overall & MEASURE_NUMBER_OF_NANS)
469 set_meta(metadata, 0, "Number of NaNs", "%f", nb_nans / (float)s->nb_channels);
470 if ((s->is_float || s->is_double) && s->measure_overall & MEASURE_NUMBER_OF_INFS)
471 set_meta(metadata, 0, "Number of Infs", "%f", nb_infs / (float)s->nb_channels);
472 if ((s->is_float || s->is_double) && s->measure_overall & MEASURE_NUMBER_OF_DENORMALS)
473 set_meta(metadata, 0, "Number of denormals", "%f", nb_denormals / (float)s->nb_channels);
476 #define UPDATE_STATS_P(type, update_func, update_float, channel_func) \
477 for (int c = 0; c < channels; c++) { \
478 ChannelStats *p = &s->chstats[c]; \
479 const type *src = (const type *)data[c]; \
480 const type * const srcend = src + samples; \
481 for (; src < srcend; src++) { \
488 #define UPDATE_STATS_I(type, update_func, update_float, channel_func) \
489 for (int c = 0; c < channels; c++) { \
490 ChannelStats *p = &s->chstats[c]; \
491 const type *src = (const type *)data[0]; \
492 const type * const srcend = src + samples * channels; \
493 for (src += c; src < srcend; src += channels) { \
500 #define UPDATE_STATS(planar, type, sample, normalizer_suffix, int_sample) \
501 if ((s->measure_overall | s->measure_perchannel) & ~MEASURE_MINMAXPEAK) { \
502 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, ); \
504 UPDATE_STATS_##planar(type, update_minmax(s, p, sample), , p->nmin = p->min normalizer_suffix; p->nmax = p->max normalizer_suffix;); \
507 static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
509 AudioStatsContext *s = inlink->dst->priv;
510 AVDictionary **metadata = &buf->metadata;
511 const int channels = s->nb_channels;
512 const int samples = buf->nb_samples;
513 const uint8_t * const * const data = (const uint8_t * const *)buf->extended_data;
515 if (s->reset_count > 0) {
516 if (s->nb_frames >= s->reset_count) {
523 switch (inlink->format) {
524 case AV_SAMPLE_FMT_DBLP:
525 UPDATE_STATS(P, double, *src, , llrint(*src * (UINT64_C(1) << 63)));
527 case AV_SAMPLE_FMT_DBL:
528 UPDATE_STATS(I, double, *src, , llrint(*src * (UINT64_C(1) << 63)));
530 case AV_SAMPLE_FMT_FLTP:
531 UPDATE_STATS(P, float, *src, , llrint(*src * (UINT64_C(1) << 31)));
533 case AV_SAMPLE_FMT_FLT:
534 UPDATE_STATS(I, float, *src, , llrint(*src * (UINT64_C(1) << 31)));
536 case AV_SAMPLE_FMT_S64P:
537 UPDATE_STATS(P, int64_t, *src, / (double)INT64_MAX, *src);
539 case AV_SAMPLE_FMT_S64:
540 UPDATE_STATS(I, int64_t, *src, / (double)INT64_MAX, *src);
542 case AV_SAMPLE_FMT_S32P:
543 UPDATE_STATS(P, int32_t, *src, / (double)INT32_MAX, *src);
545 case AV_SAMPLE_FMT_S32:
546 UPDATE_STATS(I, int32_t, *src, / (double)INT32_MAX, *src);
548 case AV_SAMPLE_FMT_S16P:
549 UPDATE_STATS(P, int16_t, *src, / (double)INT16_MAX, *src);
551 case AV_SAMPLE_FMT_S16:
552 UPDATE_STATS(I, int16_t, *src, / (double)INT16_MAX, *src);
557 set_metadata(s, metadata);
559 return ff_filter_frame(inlink->dst->outputs[0], buf);
562 static void print_stats(AVFilterContext *ctx)
564 AudioStatsContext *s = ctx->priv;
565 uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0;
566 uint64_t nb_nans = 0, nb_infs = 0, nb_denormals = 0;
567 double min_runs = 0, max_runs = 0,
568 min = DBL_MAX, max =-DBL_MAX, min_diff = DBL_MAX, max_diff = 0,
569 nmin = DBL_MAX, nmax =-DBL_MAX,
575 min_sigma_x2 = DBL_MAX,
576 max_sigma_x2 =-DBL_MAX;
580 for (c = 0; c < s->nb_channels; c++) {
581 ChannelStats *p = &s->chstats[c];
583 if (p->nb_samples < s->tc_samples)
584 p->min_sigma_x2 = p->max_sigma_x2 = p->sigma_x2 / p->nb_samples;
586 min = FFMIN(min, p->min);
587 max = FFMAX(max, p->max);
588 nmin = FFMIN(nmin, p->nmin);
589 nmax = FFMAX(nmax, p->nmax);
590 min_diff = FFMIN(min_diff, p->min_diff);
591 max_diff = FFMAX(max_diff, p->max_diff);
592 diff1_sum_x2 += p->diff1_sum_x2;
593 diff1_sum += p->diff1_sum;
594 min_sigma_x2 = FFMIN(min_sigma_x2, p->min_sigma_x2);
595 max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
596 sigma_x += p->sigma_x;
597 sigma_x2 += p->sigma_x2;
598 min_count += p->min_count;
599 max_count += p->max_count;
600 min_runs += p->min_runs;
601 max_runs += p->max_runs;
604 nb_samples += p->nb_samples;
605 nb_nans += p->nb_nans;
606 nb_infs += p->nb_infs;
607 nb_denormals += p->nb_denormals;
608 if (fabs(p->sigma_x) > fabs(max_sigma_x))
609 max_sigma_x = p->sigma_x;
611 av_log(ctx, AV_LOG_INFO, "Channel: %d\n", c + 1);
612 if (s->measure_perchannel & MEASURE_DC_OFFSET)
613 av_log(ctx, AV_LOG_INFO, "DC offset: %f\n", p->sigma_x / p->nb_samples);
614 if (s->measure_perchannel & MEASURE_MIN_LEVEL)
615 av_log(ctx, AV_LOG_INFO, "Min level: %f\n", p->min);
616 if (s->measure_perchannel & MEASURE_MAX_LEVEL)
617 av_log(ctx, AV_LOG_INFO, "Max level: %f\n", p->max);
618 if (s->measure_perchannel & MEASURE_MIN_DIFFERENCE)
619 av_log(ctx, AV_LOG_INFO, "Min difference: %f\n", p->min_diff);
620 if (s->measure_perchannel & MEASURE_MAX_DIFFERENCE)
621 av_log(ctx, AV_LOG_INFO, "Max difference: %f\n", p->max_diff);
622 if (s->measure_perchannel & MEASURE_MEAN_DIFFERENCE)
623 av_log(ctx, AV_LOG_INFO, "Mean difference: %f\n", p->diff1_sum / (p->nb_samples - 1));
624 if (s->measure_perchannel & MEASURE_RMS_DIFFERENCE)
625 av_log(ctx, AV_LOG_INFO, "RMS difference: %f\n", sqrt(p->diff1_sum_x2 / (p->nb_samples - 1)));
626 if (s->measure_perchannel & MEASURE_PEAK_LEVEL)
627 av_log(ctx, AV_LOG_INFO, "Peak level dB: %f\n", LINEAR_TO_DB(FFMAX(-p->nmin, p->nmax)));
628 if (s->measure_perchannel & MEASURE_RMS_LEVEL)
629 av_log(ctx, AV_LOG_INFO, "RMS level dB: %f\n", LINEAR_TO_DB(sqrt(p->sigma_x2 / p->nb_samples)));
630 if (s->measure_perchannel & MEASURE_RMS_PEAK)
631 av_log(ctx, AV_LOG_INFO, "RMS peak dB: %f\n", LINEAR_TO_DB(sqrt(p->max_sigma_x2)));
632 if (s->measure_perchannel & MEASURE_RMS_TROUGH)
633 if (p->min_sigma_x2 != 1)
634 av_log(ctx, AV_LOG_INFO, "RMS trough dB: %f\n",LINEAR_TO_DB(sqrt(p->min_sigma_x2)));
635 if (s->measure_perchannel & MEASURE_CREST_FACTOR)
636 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);
637 if (s->measure_perchannel & MEASURE_FLAT_FACTOR)
638 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)));
639 if (s->measure_perchannel & MEASURE_PEAK_COUNT)
640 av_log(ctx, AV_LOG_INFO, "Peak count: %"PRId64"\n", p->min_count + p->max_count);
641 if (s->measure_perchannel & MEASURE_BIT_DEPTH) {
642 bit_depth(s, p->mask, p->imask, &depth);
643 av_log(ctx, AV_LOG_INFO, "Bit depth: %u/%u\n", depth.num, depth.den);
645 if (s->measure_perchannel & MEASURE_DYNAMIC_RANGE)
646 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));
647 if (s->measure_perchannel & MEASURE_ZERO_CROSSINGS)
648 av_log(ctx, AV_LOG_INFO, "Zero crossings: %"PRId64"\n", p->zero_runs);
649 if (s->measure_perchannel & MEASURE_ZERO_CROSSINGS_RATE)
650 av_log(ctx, AV_LOG_INFO, "Zero crossings rate: %f\n", p->zero_runs/(double)p->nb_samples);
651 if ((s->is_float || s->is_double) && s->measure_perchannel & MEASURE_NUMBER_OF_NANS)
652 av_log(ctx, AV_LOG_INFO, "Number of NaNs: %"PRId64"\n", p->nb_nans);
653 if ((s->is_float || s->is_double) && s->measure_perchannel & MEASURE_NUMBER_OF_INFS)
654 av_log(ctx, AV_LOG_INFO, "Number of Infs: %"PRId64"\n", p->nb_infs);
655 if ((s->is_float || s->is_double) && s->measure_perchannel & MEASURE_NUMBER_OF_DENORMALS)
656 av_log(ctx, AV_LOG_INFO, "Number of denormals: %"PRId64"\n", p->nb_denormals);
659 av_log(ctx, AV_LOG_INFO, "Overall\n");
660 if (s->measure_overall & MEASURE_DC_OFFSET)
661 av_log(ctx, AV_LOG_INFO, "DC offset: %f\n", max_sigma_x / (nb_samples / s->nb_channels));
662 if (s->measure_overall & MEASURE_MIN_LEVEL)
663 av_log(ctx, AV_LOG_INFO, "Min level: %f\n", min);
664 if (s->measure_overall & MEASURE_MAX_LEVEL)
665 av_log(ctx, AV_LOG_INFO, "Max level: %f\n", max);
666 if (s->measure_overall & MEASURE_MIN_DIFFERENCE)
667 av_log(ctx, AV_LOG_INFO, "Min difference: %f\n", min_diff);
668 if (s->measure_overall & MEASURE_MAX_DIFFERENCE)
669 av_log(ctx, AV_LOG_INFO, "Max difference: %f\n", max_diff);
670 if (s->measure_overall & MEASURE_MEAN_DIFFERENCE)
671 av_log(ctx, AV_LOG_INFO, "Mean difference: %f\n", diff1_sum / (nb_samples - s->nb_channels));
672 if (s->measure_overall & MEASURE_RMS_DIFFERENCE)
673 av_log(ctx, AV_LOG_INFO, "RMS difference: %f\n", sqrt(diff1_sum_x2 / (nb_samples - s->nb_channels)));
674 if (s->measure_overall & MEASURE_PEAK_LEVEL)
675 av_log(ctx, AV_LOG_INFO, "Peak level dB: %f\n", LINEAR_TO_DB(FFMAX(-nmin, nmax)));
676 if (s->measure_overall & MEASURE_RMS_LEVEL)
677 av_log(ctx, AV_LOG_INFO, "RMS level dB: %f\n", LINEAR_TO_DB(sqrt(sigma_x2 / nb_samples)));
678 if (s->measure_overall & MEASURE_RMS_PEAK)
679 av_log(ctx, AV_LOG_INFO, "RMS peak dB: %f\n", LINEAR_TO_DB(sqrt(max_sigma_x2)));
680 if (s->measure_overall & MEASURE_RMS_TROUGH)
681 if (min_sigma_x2 != 1)
682 av_log(ctx, AV_LOG_INFO, "RMS trough dB: %f\n", LINEAR_TO_DB(sqrt(min_sigma_x2)));
683 if (s->measure_overall & MEASURE_FLAT_FACTOR)
684 av_log(ctx, AV_LOG_INFO, "Flat factor: %f\n", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
685 if (s->measure_overall & MEASURE_PEAK_COUNT)
686 av_log(ctx, AV_LOG_INFO, "Peak count: %f\n", (min_count + max_count) / (double)s->nb_channels);
687 if (s->measure_overall & MEASURE_BIT_DEPTH) {
688 bit_depth(s, mask, imask, &depth);
689 av_log(ctx, AV_LOG_INFO, "Bit depth: %u/%u\n", depth.num, depth.den);
691 if (s->measure_overall & MEASURE_NUMBER_OF_SAMPLES)
692 av_log(ctx, AV_LOG_INFO, "Number of samples: %"PRId64"\n", nb_samples / s->nb_channels);
693 if ((s->is_float || s->is_double) && s->measure_overall & MEASURE_NUMBER_OF_NANS)
694 av_log(ctx, AV_LOG_INFO, "Number of NaNs: %f\n", nb_nans / (float)s->nb_channels);
695 if ((s->is_float || s->is_double) && s->measure_overall & MEASURE_NUMBER_OF_INFS)
696 av_log(ctx, AV_LOG_INFO, "Number of Infs: %f\n", nb_infs / (float)s->nb_channels);
697 if ((s->is_float || s->is_double) && s->measure_overall & MEASURE_NUMBER_OF_DENORMALS)
698 av_log(ctx, AV_LOG_INFO, "Number of denormals: %f\n", nb_denormals / (float)s->nb_channels);
701 static av_cold void uninit(AVFilterContext *ctx)
703 AudioStatsContext *s = ctx->priv;
707 av_freep(&s->chstats);
710 static const AVFilterPad astats_inputs[] = {
713 .type = AVMEDIA_TYPE_AUDIO,
714 .filter_frame = filter_frame,
719 static const AVFilterPad astats_outputs[] = {
722 .type = AVMEDIA_TYPE_AUDIO,
723 .config_props = config_output,
728 AVFilter ff_af_astats = {
730 .description = NULL_IF_CONFIG_SMALL("Show time domain statistics about audio frames."),
731 .query_formats = query_formats,
732 .priv_size = sizeof(AudioStatsContext),
733 .priv_class = &astats_class,
735 .inputs = astats_inputs,
736 .outputs = astats_outputs,