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;
34 double min_run, max_run;
35 double min_runs, max_runs;
36 double min_diff, max_diff;
39 uint64_t min_count, max_count;
45 ChannelStats *chstats;
55 #define OFFSET(x) offsetof(AudioStatsContext, x)
56 #define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
58 static const AVOption astats_options[] = {
59 { "length", "set the window length", OFFSET(time_constant), AV_OPT_TYPE_DOUBLE, {.dbl=.05}, .01, 10, FLAGS },
60 { "metadata", "inject metadata in the filtergraph", OFFSET(metadata), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS },
61 { "reset", "recalculate stats after this many frames", OFFSET(reset_count), AV_OPT_TYPE_INT, {.i64=0}, 0, INT_MAX, FLAGS },
65 AVFILTER_DEFINE_CLASS(astats);
67 static int query_formats(AVFilterContext *ctx)
69 AVFilterFormats *formats;
70 AVFilterChannelLayouts *layouts;
71 static const enum AVSampleFormat sample_fmts[] = {
72 AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_DBLP,
77 layouts = ff_all_channel_counts();
79 return AVERROR(ENOMEM);
80 ret = ff_set_common_channel_layouts(ctx, layouts);
84 formats = ff_make_format_list(sample_fmts);
86 return AVERROR(ENOMEM);
87 ret = ff_set_common_formats(ctx, formats);
91 formats = ff_all_samplerates();
93 return AVERROR(ENOMEM);
94 return ff_set_common_samplerates(ctx, formats);
97 static void reset_stats(AudioStatsContext *s)
101 for (c = 0; c < s->nb_channels; c++) {
102 ChannelStats *p = &s->chstats[c];
104 p->min = p->min_sigma_x2 = DBL_MAX;
105 p->max = p->max_sigma_x2 = DBL_MIN;
106 p->min_diff = DBL_MAX;
107 p->max_diff = DBL_MIN;
119 p->imask = 0xFFFFFFFFFFFFFFFF;
126 static int config_output(AVFilterLink *outlink)
128 AudioStatsContext *s = outlink->src->priv;
130 s->chstats = av_calloc(sizeof(*s->chstats), outlink->channels);
132 return AVERROR(ENOMEM);
133 s->nb_channels = outlink->channels;
134 s->mult = exp((-1 / s->time_constant / outlink->sample_rate));
135 s->tc_samples = 5 * s->time_constant * outlink->sample_rate + .5;
143 static void bit_depth(uint64_t mask, uint64_t imask, AVRational *depth)
145 unsigned result = 64;
147 mask = mask & (~imask);
149 for (; result && !(mask & 1); --result, mask >>= 1);
154 for (; result; --result, mask >>= 1)
159 static inline void update_stat(AudioStatsContext *s, ChannelStats *p, double d)
166 } else if (d == p->min) {
168 p->min_run = d == p->last ? p->min_run + 1 : 1;
169 } else if (p->last == p->min) {
170 p->min_runs += p->min_run * p->min_run;
178 } else if (d == p->max) {
180 p->max_run = d == p->last ? p->max_run + 1 : 1;
181 } else if (p->last == p->max) {
182 p->max_runs += p->max_run * p->max_run;
186 p->sigma_x2 += d * d;
187 p->avg_sigma_x2 = p->avg_sigma_x2 * s->mult + (1.0 - s->mult) * d * d;
188 p->min_diff = FFMIN(p->min_diff, fabs(d - p->last));
189 p->max_diff = FFMAX(p->max_diff, fabs(d - p->last));
190 p->diff1_sum += fabs(d - p->last);
192 p->mask |= llrint(d * (UINT64_C(1) << 63));
193 p->imask &= llrint(d * (UINT64_C(1) << 63));
195 if (p->nb_samples >= s->tc_samples) {
196 p->max_sigma_x2 = FFMAX(p->max_sigma_x2, p->avg_sigma_x2);
197 p->min_sigma_x2 = FFMIN(p->min_sigma_x2, p->avg_sigma_x2);
202 static void set_meta(AVDictionary **metadata, int chan, const char *key,
203 const char *fmt, double val)
208 snprintf(value, sizeof(value), fmt, val);
210 snprintf(key2, sizeof(key2), "lavfi.astats.%d.%s", chan, key);
212 snprintf(key2, sizeof(key2), "lavfi.astats.%s", key);
213 av_dict_set(metadata, key2, value, 0);
216 #define LINEAR_TO_DB(x) (log10(x) * 20)
218 static void set_metadata(AudioStatsContext *s, AVDictionary **metadata)
220 uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0;
221 double min_runs = 0, max_runs = 0,
222 min = DBL_MAX, max = DBL_MIN, min_diff = DBL_MAX, max_diff = 0,
227 min_sigma_x2 = DBL_MAX,
228 max_sigma_x2 = DBL_MIN;
232 for (c = 0; c < s->nb_channels; c++) {
233 ChannelStats *p = &s->chstats[c];
235 if (p->nb_samples < s->tc_samples)
236 p->min_sigma_x2 = p->max_sigma_x2 = p->sigma_x2 / p->nb_samples;
238 min = FFMIN(min, p->min);
239 max = FFMAX(max, p->max);
240 min_diff = FFMIN(min_diff, p->min_diff);
241 max_diff = FFMAX(max_diff, p->max_diff);
242 diff1_sum += p->diff1_sum,
243 min_sigma_x2 = FFMIN(min_sigma_x2, p->min_sigma_x2);
244 max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
245 sigma_x += p->sigma_x;
246 sigma_x2 += p->sigma_x2;
247 min_count += p->min_count;
248 max_count += p->max_count;
249 min_runs += p->min_runs;
250 max_runs += p->max_runs;
253 nb_samples += p->nb_samples;
254 if (fabs(p->sigma_x) > fabs(max_sigma_x))
255 max_sigma_x = p->sigma_x;
257 set_meta(metadata, c + 1, "DC_offset", "%f", p->sigma_x / p->nb_samples);
258 set_meta(metadata, c + 1, "Min_level", "%f", p->min);
259 set_meta(metadata, c + 1, "Max_level", "%f", p->max);
260 set_meta(metadata, c + 1, "Min_difference", "%f", p->min_diff);
261 set_meta(metadata, c + 1, "Max_difference", "%f", p->max_diff);
262 set_meta(metadata, c + 1, "Mean_difference", "%f", p->diff1_sum / (p->nb_samples - 1));
263 set_meta(metadata, c + 1, "Peak_level", "%f", LINEAR_TO_DB(FFMAX(-p->min, p->max)));
264 set_meta(metadata, c + 1, "RMS_level", "%f", LINEAR_TO_DB(sqrt(p->sigma_x2 / p->nb_samples)));
265 set_meta(metadata, c + 1, "RMS_peak", "%f", LINEAR_TO_DB(sqrt(p->max_sigma_x2)));
266 set_meta(metadata, c + 1, "RMS_trough", "%f", LINEAR_TO_DB(sqrt(p->min_sigma_x2)));
267 set_meta(metadata, c + 1, "Crest_factor", "%f", p->sigma_x2 ? FFMAX(-p->min, p->max) / sqrt(p->sigma_x2 / p->nb_samples) : 1);
268 set_meta(metadata, c + 1, "Flat_factor", "%f", LINEAR_TO_DB((p->min_runs + p->max_runs) / (p->min_count + p->max_count)));
269 set_meta(metadata, c + 1, "Peak_count", "%f", (float)(p->min_count + p->max_count));
270 bit_depth(p->mask, p->imask, &depth);
271 set_meta(metadata, c + 1, "Bit_depth", "%f", depth.num);
272 set_meta(metadata, c + 1, "Bit_depth2", "%f", depth.den);
275 set_meta(metadata, 0, "Overall.DC_offset", "%f", max_sigma_x / (nb_samples / s->nb_channels));
276 set_meta(metadata, 0, "Overall.Min_level", "%f", min);
277 set_meta(metadata, 0, "Overall.Max_level", "%f", max);
278 set_meta(metadata, 0, "Overall.Min_difference", "%f", min_diff);
279 set_meta(metadata, 0, "Overall.Max_difference", "%f", max_diff);
280 set_meta(metadata, 0, "Overall.Mean_difference", "%f", diff1_sum / (nb_samples - s->nb_channels));
281 set_meta(metadata, 0, "Overall.Peak_level", "%f", LINEAR_TO_DB(FFMAX(-min, max)));
282 set_meta(metadata, 0, "Overall.RMS_level", "%f", LINEAR_TO_DB(sqrt(sigma_x2 / nb_samples)));
283 set_meta(metadata, 0, "Overall.RMS_peak", "%f", LINEAR_TO_DB(sqrt(max_sigma_x2)));
284 set_meta(metadata, 0, "Overall.RMS_trough", "%f", LINEAR_TO_DB(sqrt(min_sigma_x2)));
285 set_meta(metadata, 0, "Overall.Flat_factor", "%f", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
286 set_meta(metadata, 0, "Overall.Peak_count", "%f", (float)(min_count + max_count) / (double)s->nb_channels);
287 bit_depth(mask, imask, &depth);
288 set_meta(metadata, 0, "Overall.Bit_depth", "%f", depth.num);
289 set_meta(metadata, 0, "Overall.Bit_depth2", "%f", depth.den);
290 set_meta(metadata, 0, "Overall.Number_of_samples", "%f", nb_samples / s->nb_channels);
293 static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
295 AudioStatsContext *s = inlink->dst->priv;
296 AVDictionary **metadata = avpriv_frame_get_metadatap(buf);
297 const int channels = s->nb_channels;
301 if (s->reset_count > 0) {
302 if (s->nb_frames >= s->reset_count) {
309 switch (inlink->format) {
310 case AV_SAMPLE_FMT_DBLP:
311 for (c = 0; c < channels; c++) {
312 ChannelStats *p = &s->chstats[c];
313 src = (const double *)buf->extended_data[c];
315 for (i = 0; i < buf->nb_samples; i++, src++)
316 update_stat(s, p, *src);
319 case AV_SAMPLE_FMT_DBL:
320 src = (const double *)buf->extended_data[0];
322 for (i = 0; i < buf->nb_samples; i++) {
323 for (c = 0; c < channels; c++, src++)
324 update_stat(s, &s->chstats[c], *src);
330 set_metadata(s, metadata);
332 return ff_filter_frame(inlink->dst->outputs[0], buf);
335 static void print_stats(AVFilterContext *ctx)
337 AudioStatsContext *s = ctx->priv;
338 uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0;
339 double min_runs = 0, max_runs = 0,
340 min = DBL_MAX, max = DBL_MIN, min_diff = DBL_MAX, max_diff = 0,
345 min_sigma_x2 = DBL_MAX,
346 max_sigma_x2 = DBL_MIN;
350 for (c = 0; c < s->nb_channels; c++) {
351 ChannelStats *p = &s->chstats[c];
353 if (p->nb_samples < s->tc_samples)
354 p->min_sigma_x2 = p->max_sigma_x2 = p->sigma_x2 / p->nb_samples;
356 min = FFMIN(min, p->min);
357 max = FFMAX(max, p->max);
358 min_diff = FFMIN(min_diff, p->min_diff);
359 max_diff = FFMAX(max_diff, p->max_diff);
360 diff1_sum += p->diff1_sum,
361 min_sigma_x2 = FFMIN(min_sigma_x2, p->min_sigma_x2);
362 max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
363 sigma_x += p->sigma_x;
364 sigma_x2 += p->sigma_x2;
365 min_count += p->min_count;
366 max_count += p->max_count;
367 min_runs += p->min_runs;
368 max_runs += p->max_runs;
371 nb_samples += p->nb_samples;
372 if (fabs(p->sigma_x) > fabs(max_sigma_x))
373 max_sigma_x = p->sigma_x;
375 av_log(ctx, AV_LOG_INFO, "Channel: %d\n", c + 1);
376 av_log(ctx, AV_LOG_INFO, "DC offset: %f\n", p->sigma_x / p->nb_samples);
377 av_log(ctx, AV_LOG_INFO, "Min level: %f\n", p->min);
378 av_log(ctx, AV_LOG_INFO, "Max level: %f\n", p->max);
379 av_log(ctx, AV_LOG_INFO, "Min difference: %f\n", p->min_diff);
380 av_log(ctx, AV_LOG_INFO, "Max difference: %f\n", p->max_diff);
381 av_log(ctx, AV_LOG_INFO, "Mean difference: %f\n", p->diff1_sum / (p->nb_samples - 1));
382 av_log(ctx, AV_LOG_INFO, "Peak level dB: %f\n", LINEAR_TO_DB(FFMAX(-p->min, p->max)));
383 av_log(ctx, AV_LOG_INFO, "RMS level dB: %f\n", LINEAR_TO_DB(sqrt(p->sigma_x2 / p->nb_samples)));
384 av_log(ctx, AV_LOG_INFO, "RMS peak dB: %f\n", LINEAR_TO_DB(sqrt(p->max_sigma_x2)));
385 if (p->min_sigma_x2 != 1)
386 av_log(ctx, AV_LOG_INFO, "RMS trough dB: %f\n",LINEAR_TO_DB(sqrt(p->min_sigma_x2)));
387 av_log(ctx, AV_LOG_INFO, "Crest factor: %f\n", p->sigma_x2 ? FFMAX(-p->min, p->max) / sqrt(p->sigma_x2 / p->nb_samples) : 1);
388 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)));
389 av_log(ctx, AV_LOG_INFO, "Peak count: %"PRId64"\n", p->min_count + p->max_count);
390 bit_depth(p->mask, p->imask, &depth);
391 av_log(ctx, AV_LOG_INFO, "Bit depth: %u/%u\n", depth.num, depth.den);
394 av_log(ctx, AV_LOG_INFO, "Overall\n");
395 av_log(ctx, AV_LOG_INFO, "DC offset: %f\n", max_sigma_x / (nb_samples / s->nb_channels));
396 av_log(ctx, AV_LOG_INFO, "Min level: %f\n", min);
397 av_log(ctx, AV_LOG_INFO, "Max level: %f\n", max);
398 av_log(ctx, AV_LOG_INFO, "Min difference: %f\n", min_diff);
399 av_log(ctx, AV_LOG_INFO, "Max difference: %f\n", max_diff);
400 av_log(ctx, AV_LOG_INFO, "Mean difference: %f\n", diff1_sum / (nb_samples - s->nb_channels));
401 av_log(ctx, AV_LOG_INFO, "Peak level dB: %f\n", LINEAR_TO_DB(FFMAX(-min, max)));
402 av_log(ctx, AV_LOG_INFO, "RMS level dB: %f\n", LINEAR_TO_DB(sqrt(sigma_x2 / nb_samples)));
403 av_log(ctx, AV_LOG_INFO, "RMS peak dB: %f\n", LINEAR_TO_DB(sqrt(max_sigma_x2)));
404 if (min_sigma_x2 != 1)
405 av_log(ctx, AV_LOG_INFO, "RMS trough dB: %f\n", LINEAR_TO_DB(sqrt(min_sigma_x2)));
406 av_log(ctx, AV_LOG_INFO, "Flat factor: %f\n", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
407 av_log(ctx, AV_LOG_INFO, "Peak count: %f\n", (min_count + max_count) / (double)s->nb_channels);
408 bit_depth(mask, imask, &depth);
409 av_log(ctx, AV_LOG_INFO, "Bit depth: %u/%u\n", depth.num, depth.den);
410 av_log(ctx, AV_LOG_INFO, "Number of samples: %"PRId64"\n", nb_samples / s->nb_channels);
413 static av_cold void uninit(AVFilterContext *ctx)
415 AudioStatsContext *s = ctx->priv;
419 av_freep(&s->chstats);
422 static const AVFilterPad astats_inputs[] = {
425 .type = AVMEDIA_TYPE_AUDIO,
426 .filter_frame = filter_frame,
431 static const AVFilterPad astats_outputs[] = {
434 .type = AVMEDIA_TYPE_AUDIO,
435 .config_props = config_output,
440 AVFilter ff_af_astats = {
442 .description = NULL_IF_CONFIG_SMALL("Show time domain statistics about audio frames."),
443 .query_formats = query_formats,
444 .priv_size = sizeof(AudioStatsContext),
445 .priv_class = &astats_class,
447 .inputs = astats_inputs,
448 .outputs = astats_outputs,