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 uint64_t min_count, max_count;
42 ChannelStats *chstats;
52 #define OFFSET(x) offsetof(AudioStatsContext, x)
53 #define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
55 static const AVOption astats_options[] = {
56 { "length", "set the window length", OFFSET(time_constant), AV_OPT_TYPE_DOUBLE, {.dbl=.05}, .01, 10, FLAGS },
57 { "metadata", "inject metadata in the filtergraph", OFFSET(metadata), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, FLAGS },
58 { "reset", "recalculate stats after this many frames", OFFSET(reset_count), AV_OPT_TYPE_INT, {.i64=0}, 0, INT_MAX, FLAGS },
62 AVFILTER_DEFINE_CLASS(astats);
64 static int query_formats(AVFilterContext *ctx)
66 AVFilterFormats *formats;
67 AVFilterChannelLayouts *layouts;
68 static const enum AVSampleFormat sample_fmts[] = {
69 AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_DBLP,
74 layouts = ff_all_channel_layouts();
76 return AVERROR(ENOMEM);
77 ret = ff_set_common_channel_layouts(ctx, layouts);
81 formats = ff_make_format_list(sample_fmts);
83 return AVERROR(ENOMEM);
84 ret = ff_set_common_formats(ctx, formats);
88 formats = ff_all_samplerates();
90 return AVERROR(ENOMEM);
91 return ff_set_common_samplerates(ctx, formats);
94 static void reset_stats(AudioStatsContext *s)
98 memset(s->chstats, 0, sizeof(*s->chstats));
100 for (c = 0; c < s->nb_channels; c++) {
101 ChannelStats *p = &s->chstats[c];
103 p->min = p->min_sigma_x2 = DBL_MAX;
104 p->max = p->max_sigma_x2 = DBL_MIN;
108 static int config_output(AVFilterLink *outlink)
110 AudioStatsContext *s = outlink->src->priv;
112 s->chstats = av_calloc(sizeof(*s->chstats), outlink->channels);
114 return AVERROR(ENOMEM);
115 s->nb_channels = outlink->channels;
116 s->mult = exp((-1 / s->time_constant / outlink->sample_rate));
117 s->tc_samples = 5 * s->time_constant * outlink->sample_rate + .5;
124 static inline void update_stat(AudioStatsContext *s, ChannelStats *p, double d)
131 } else if (d == p->min) {
133 p->min_run = d == p->last ? p->min_run + 1 : 1;
134 } else if (p->last == p->min) {
135 p->min_runs += p->min_run * p->min_run;
143 } else if (d == p->max) {
145 p->max_run = d == p->last ? p->max_run + 1 : 1;
146 } else if (p->last == p->max) {
147 p->max_runs += p->max_run * p->max_run;
151 p->sigma_x2 += d * d;
152 p->avg_sigma_x2 = p->avg_sigma_x2 * s->mult + (1.0 - s->mult) * d * d;
155 if (p->nb_samples >= s->tc_samples) {
156 p->max_sigma_x2 = FFMAX(p->max_sigma_x2, p->avg_sigma_x2);
157 p->min_sigma_x2 = FFMIN(p->min_sigma_x2, p->avg_sigma_x2);
162 static void set_meta(AVDictionary **metadata, int chan, const char *key,
163 const char *fmt, double val)
168 snprintf(value, sizeof(value), fmt, val);
170 snprintf(key2, sizeof(key2), "lavfi.astats.%d.%s", chan, key);
172 snprintf(key2, sizeof(key2), "lavfi.astats.%s", key);
173 av_dict_set(metadata, key2, value, 0);
176 #define LINEAR_TO_DB(x) (log10(x) * 20)
178 static void set_metadata(AudioStatsContext *s, AVDictionary **metadata)
180 uint64_t min_count = 0, max_count = 0, nb_samples = 0;
181 double min_runs = 0, max_runs = 0,
182 min = DBL_MAX, max = DBL_MIN,
186 min_sigma_x2 = DBL_MAX,
187 max_sigma_x2 = DBL_MIN;
190 for (c = 0; c < s->nb_channels; c++) {
191 ChannelStats *p = &s->chstats[c];
193 if (p->nb_samples < s->tc_samples)
194 p->min_sigma_x2 = p->max_sigma_x2 = p->sigma_x2 / p->nb_samples;
196 min = FFMIN(min, p->min);
197 max = FFMAX(max, p->max);
198 min_sigma_x2 = FFMIN(min_sigma_x2, p->min_sigma_x2);
199 max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
200 sigma_x += p->sigma_x;
201 sigma_x2 += p->sigma_x2;
202 min_count += p->min_count;
203 max_count += p->max_count;
204 min_runs += p->min_runs;
205 max_runs += p->max_runs;
206 nb_samples += p->nb_samples;
207 if (fabs(p->sigma_x) > fabs(max_sigma_x))
208 max_sigma_x = p->sigma_x;
210 set_meta(metadata, c + 1, "DC_offset", "%f", p->sigma_x / p->nb_samples);
211 set_meta(metadata, c + 1, "Min_level", "%f", p->min);
212 set_meta(metadata, c + 1, "Max_level", "%f", p->max);
213 set_meta(metadata, c + 1, "Peak_level", "%f", LINEAR_TO_DB(FFMAX(-p->min, p->max)));
214 set_meta(metadata, c + 1, "RMS_level", "%f", LINEAR_TO_DB(sqrt(p->sigma_x2 / p->nb_samples)));
215 set_meta(metadata, c + 1, "RMS_peak", "%f", LINEAR_TO_DB(sqrt(p->max_sigma_x2)));
216 set_meta(metadata, c + 1, "RMS_trough", "%f", LINEAR_TO_DB(sqrt(p->min_sigma_x2)));
217 set_meta(metadata, c + 1, "Crest_factor", "%f", p->sigma_x2 ? FFMAX(-p->min, p->max) / sqrt(p->sigma_x2 / p->nb_samples) : 1);
218 set_meta(metadata, c + 1, "Flat_factor", "%f", LINEAR_TO_DB((p->min_runs + p->max_runs) / (p->min_count + p->max_count)));
219 set_meta(metadata, c + 1, "Peak_count", "%f", (float)(p->min_count + p->max_count));
222 set_meta(metadata, 0, "Overall.DC_offset", "%f", max_sigma_x / (nb_samples / s->nb_channels));
223 set_meta(metadata, 0, "Overall.Min_level", "%f", min);
224 set_meta(metadata, 0, "Overall.Max_level", "%f", max);
225 set_meta(metadata, 0, "Overall.Peak_level", "%f", LINEAR_TO_DB(FFMAX(-min, max)));
226 set_meta(metadata, 0, "Overall.RMS_level", "%f", LINEAR_TO_DB(sqrt(sigma_x2 / nb_samples)));
227 set_meta(metadata, 0, "Overall.RMS_peak", "%f", LINEAR_TO_DB(sqrt(max_sigma_x2)));
228 set_meta(metadata, 0, "Overall.RMS_trough", "%f", LINEAR_TO_DB(sqrt(min_sigma_x2)));
229 set_meta(metadata, 0, "Overall.Flat_factor", "%f", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
230 set_meta(metadata, 0, "Overall.Peak_count", "%f", (float)(min_count + max_count) / (double)s->nb_channels);
231 set_meta(metadata, 0, "Overall.Number_of_samples", "%f", nb_samples / s->nb_channels);
234 static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
236 AudioStatsContext *s = inlink->dst->priv;
237 AVDictionary **metadata = avpriv_frame_get_metadatap(buf);
238 const int channels = s->nb_channels;
242 switch (inlink->format) {
243 case AV_SAMPLE_FMT_DBLP:
244 for (c = 0; c < channels; c++) {
245 ChannelStats *p = &s->chstats[c];
246 src = (const double *)buf->extended_data[c];
248 for (i = 0; i < buf->nb_samples; i++, src++)
249 update_stat(s, p, *src);
252 case AV_SAMPLE_FMT_DBL:
253 src = (const double *)buf->extended_data[0];
255 for (i = 0; i < buf->nb_samples; i++) {
256 for (c = 0; c < channels; c++, src++)
257 update_stat(s, &s->chstats[c], *src);
263 set_metadata(s, metadata);
265 if (s->reset_count > 0) {
267 if (s->nb_frames >= s->reset_count) {
273 return ff_filter_frame(inlink->dst->outputs[0], buf);
276 static void print_stats(AVFilterContext *ctx)
278 AudioStatsContext *s = ctx->priv;
279 uint64_t min_count = 0, max_count = 0, nb_samples = 0;
280 double min_runs = 0, max_runs = 0,
281 min = DBL_MAX, max = DBL_MIN,
285 min_sigma_x2 = DBL_MAX,
286 max_sigma_x2 = DBL_MIN;
289 for (c = 0; c < s->nb_channels; c++) {
290 ChannelStats *p = &s->chstats[c];
292 if (p->nb_samples < s->tc_samples)
293 p->min_sigma_x2 = p->max_sigma_x2 = p->sigma_x2 / p->nb_samples;
295 min = FFMIN(min, p->min);
296 max = FFMAX(max, p->max);
297 min_sigma_x2 = FFMIN(min_sigma_x2, p->min_sigma_x2);
298 max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
299 sigma_x += p->sigma_x;
300 sigma_x2 += p->sigma_x2;
301 min_count += p->min_count;
302 max_count += p->max_count;
303 min_runs += p->min_runs;
304 max_runs += p->max_runs;
305 nb_samples += p->nb_samples;
306 if (fabs(p->sigma_x) > fabs(max_sigma_x))
307 max_sigma_x = p->sigma_x;
309 av_log(ctx, AV_LOG_INFO, "Channel: %d\n", c + 1);
310 av_log(ctx, AV_LOG_INFO, "DC offset: %f\n", p->sigma_x / p->nb_samples);
311 av_log(ctx, AV_LOG_INFO, "Min level: %f\n", p->min);
312 av_log(ctx, AV_LOG_INFO, "Max level: %f\n", p->max);
313 av_log(ctx, AV_LOG_INFO, "Peak level dB: %f\n", LINEAR_TO_DB(FFMAX(-p->min, p->max)));
314 av_log(ctx, AV_LOG_INFO, "RMS level dB: %f\n", LINEAR_TO_DB(sqrt(p->sigma_x2 / p->nb_samples)));
315 av_log(ctx, AV_LOG_INFO, "RMS peak dB: %f\n", LINEAR_TO_DB(sqrt(p->max_sigma_x2)));
316 if (p->min_sigma_x2 != 1)
317 av_log(ctx, AV_LOG_INFO, "RMS trough dB: %f\n",LINEAR_TO_DB(sqrt(p->min_sigma_x2)));
318 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);
319 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)));
320 av_log(ctx, AV_LOG_INFO, "Peak count: %"PRId64"\n", p->min_count + p->max_count);
323 av_log(ctx, AV_LOG_INFO, "Overall\n");
324 av_log(ctx, AV_LOG_INFO, "DC offset: %f\n", max_sigma_x / (nb_samples / s->nb_channels));
325 av_log(ctx, AV_LOG_INFO, "Min level: %f\n", min);
326 av_log(ctx, AV_LOG_INFO, "Max level: %f\n", max);
327 av_log(ctx, AV_LOG_INFO, "Peak level dB: %f\n", LINEAR_TO_DB(FFMAX(-min, max)));
328 av_log(ctx, AV_LOG_INFO, "RMS level dB: %f\n", LINEAR_TO_DB(sqrt(sigma_x2 / nb_samples)));
329 av_log(ctx, AV_LOG_INFO, "RMS peak dB: %f\n", LINEAR_TO_DB(sqrt(max_sigma_x2)));
330 if (min_sigma_x2 != 1)
331 av_log(ctx, AV_LOG_INFO, "RMS trough dB: %f\n", LINEAR_TO_DB(sqrt(min_sigma_x2)));
332 av_log(ctx, AV_LOG_INFO, "Flat factor: %f\n", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
333 av_log(ctx, AV_LOG_INFO, "Peak count: %f\n", (min_count + max_count) / (double)s->nb_channels);
334 av_log(ctx, AV_LOG_INFO, "Number of samples: %"PRId64"\n", nb_samples / s->nb_channels);
337 static av_cold void uninit(AVFilterContext *ctx)
339 AudioStatsContext *s = ctx->priv;
343 av_freep(&s->chstats);
346 static const AVFilterPad astats_inputs[] = {
349 .type = AVMEDIA_TYPE_AUDIO,
350 .filter_frame = filter_frame,
355 static const AVFilterPad astats_outputs[] = {
358 .type = AVMEDIA_TYPE_AUDIO,
359 .config_props = config_output,
364 AVFilter ff_af_astats = {
366 .description = NULL_IF_CONFIG_SMALL("Show time domain statistics about audio frames."),
367 .query_formats = query_formats,
368 .priv_size = sizeof(AudioStatsContext),
369 .priv_class = &astats_class,
371 .inputs = astats_inputs,
372 .outputs = astats_outputs,