#include "avfilter.h"
#include "internal.h"
+#define HISTOGRAM_SIZE 8192
+#define HISTOGRAM_MAX (HISTOGRAM_SIZE-1)
+
#define MEASURE_ALL UINT_MAX
#define MEASURE_NONE 0
#define MEASURE_NUMBER_OF_NANS (1 << 19)
#define MEASURE_NUMBER_OF_INFS (1 << 20)
#define MEASURE_NUMBER_OF_DENORMALS (1 << 21)
+#define MEASURE_NOISE_FLOOR (1 << 22)
+#define MEASURE_NOISE_FLOOR_COUNT (1 << 23)
#define MEASURE_MINMAXPEAK (MEASURE_MIN_LEVEL | MEASURE_MAX_LEVEL | MEASURE_PEAK_LEVEL)
double diff1_sum_x2;
uint64_t mask, imask;
uint64_t min_count, max_count;
+ uint64_t noise_floor_count;
uint64_t zero_runs;
uint64_t nb_samples;
uint64_t nb_nans;
uint64_t nb_infs;
uint64_t nb_denormals;
+ double *win_samples;
+ unsigned histogram[HISTOGRAM_SIZE];
+ int win_pos;
+ int max_index;
+ double noise_floor;
} ChannelStats;
typedef struct AudioStatsContext {
{ "Dynamic_range" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_DYNAMIC_RANGE }, 0, 0, FLAGS, "measure" },
{ "Zero_crossings" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_ZERO_CROSSINGS }, 0, 0, FLAGS, "measure" },
{ "Zero_crossings_rate" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_ZERO_CROSSINGS_RATE }, 0, 0, FLAGS, "measure" },
+ { "Noise_floor" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_NOISE_FLOOR }, 0, 0, FLAGS, "measure" },
+ { "Noise_floor_count" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_NOISE_FLOOR_COUNT }, 0, 0, FLAGS, "measure" },
{ "Number_of_samples" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_NUMBER_OF_SAMPLES }, 0, 0, FLAGS, "measure" },
{ "Number_of_NaNs" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_NUMBER_OF_NANS }, 0, 0, FLAGS, "measure" },
{ "Number_of_Infs" , "", 0, AV_OPT_TYPE_CONST, {.i64=MEASURE_NUMBER_OF_INFS }, 0, 0, FLAGS, "measure" },
p->nb_nans = 0;
p->nb_infs = 0;
p->nb_denormals = 0;
+ p->last = NAN;
+ p->noise_floor = NAN;
+ p->noise_floor_count = 0;
+ p->win_pos = 0;
+ memset(p->win_samples, 0, s->tc_samples * sizeof(*p->win_samples));
+ memset(p->histogram, 0, sizeof(p->histogram));
}
}
s->chstats = av_calloc(sizeof(*s->chstats), outlink->channels);
if (!s->chstats)
return AVERROR(ENOMEM);
+
+ s->tc_samples = 5 * s->time_constant * outlink->sample_rate + .5;
s->nb_channels = outlink->channels;
+
+ for (int i = 0; i < s->nb_channels; i++) {
+ ChannelStats *p = &s->chstats[i];
+
+ p->win_samples = av_calloc(s->tc_samples, sizeof(*p->win_samples));
+ if (!p->win_samples)
+ return AVERROR(ENOMEM);
+ }
+
s->mult = exp((-1 / s->time_constant / outlink->sample_rate));
- s->tc_samples = 5 * s->time_constant * outlink->sample_rate + .5;
s->nb_frames = 0;
s->maxbitdepth = av_get_bytes_per_sample(outlink->format) * 8;
s->is_double = outlink->format == AV_SAMPLE_FMT_DBL ||
static inline void update_stat(AudioStatsContext *s, ChannelStats *p, double d, double nd, int64_t i)
{
+ double drop;
+ int index;
+
if (d < p->min) {
p->min = d;
p->nmin = nd;
p->sigma_x += nd;
p->sigma_x2 += nd * nd;
p->avg_sigma_x2 = p->avg_sigma_x2 * s->mult + (1.0 - s->mult) * nd * nd;
- p->min_diff = FFMIN(p->min_diff, fabs(d - p->last));
- p->max_diff = FFMAX(p->max_diff, fabs(d - p->last));
- p->diff1_sum += fabs(d - p->last);
- p->diff1_sum_x2 += (d - p->last) * (d - p->last);
+ if (!isnan(p->last)) {
+ p->min_diff = FFMIN(p->min_diff, fabs(d - p->last));
+ p->max_diff = FFMAX(p->max_diff, fabs(d - p->last));
+ p->diff1_sum += fabs(d - p->last);
+ p->diff1_sum_x2 += (d - p->last) * (d - p->last);
+ }
p->last = d;
p->mask |= i;
p->imask &= i;
+ drop = p->win_samples[p->win_pos];
+ p->win_samples[p->win_pos] = nd;
+ index = av_clip(lrint(av_clipd(FFABS(nd), 0.0, 1.0) * HISTOGRAM_MAX), 0, HISTOGRAM_MAX);
+ p->max_index = FFMAX(p->max_index, index);
+ p->histogram[index]++;
+ if (!isnan(p->noise_floor))
+ p->histogram[av_clip(lrint(av_clipd(FFABS(drop), 0.0, 1.0) * HISTOGRAM_MAX), 0, HISTOGRAM_MAX)]--;
+ p->win_pos++;
+
+ while (p->histogram[p->max_index] == 0)
+ p->max_index--;
+ if (p->win_pos >= s->tc_samples || !isnan(p->noise_floor)) {
+ double noise_floor = 1.;
+
+ for (int i = p->max_index; i >= 0; i--) {
+ if (p->histogram[i]) {
+ noise_floor = i / (double)HISTOGRAM_MAX;
+ break;
+ }
+ }
+
+ if (isnan(p->noise_floor)) {
+ p->noise_floor = noise_floor;
+ p->noise_floor_count = 1;
+ } else {
+ if (noise_floor < p->noise_floor) {
+ p->noise_floor = noise_floor;
+ p->noise_floor_count = 1;
+ } else if (noise_floor == p->noise_floor) {
+ p->noise_floor_count++;
+ }
+ }
+ }
+
+ if (p->win_pos >= s->tc_samples) {
+ p->win_pos = 0;
+ }
+
if (p->nb_samples >= s->tc_samples) {
p->max_sigma_x2 = FFMAX(p->max_sigma_x2, p->avg_sigma_x2);
p->min_sigma_x2 = FFMIN(p->min_sigma_x2, p->avg_sigma_x2);
static void set_metadata(AudioStatsContext *s, AVDictionary **metadata)
{
- uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0;
+ uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0, noise_floor_count = 0;
uint64_t nb_nans = 0, nb_infs = 0, nb_denormals = 0;
double min_runs = 0, max_runs = 0,
min = DBL_MAX, max =-DBL_MAX, min_diff = DBL_MAX, max_diff = 0,
diff1_sum_x2 = 0,
sigma_x = 0,
sigma_x2 = 0,
+ noise_floor = 0,
min_sigma_x2 = DBL_MAX,
max_sigma_x2 =-DBL_MAX;
AVRational depth;
max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
sigma_x += p->sigma_x;
sigma_x2 += p->sigma_x2;
+ noise_floor = FFMAX(noise_floor, p->noise_floor);
+ noise_floor_count += p->noise_floor_count;
min_count += p->min_count;
max_count += p->max_count;
min_runs += p->min_runs;
set_meta(metadata, c + 1, "Flat_factor", "%f", LINEAR_TO_DB((p->min_runs + p->max_runs) / (p->min_count + p->max_count)));
if (s->measure_perchannel & MEASURE_PEAK_COUNT)
set_meta(metadata, c + 1, "Peak_count", "%f", (float)(p->min_count + p->max_count));
+ if (s->measure_perchannel & MEASURE_NOISE_FLOOR)
+ set_meta(metadata, c + 1, "Noise_floor", "%f", LINEAR_TO_DB(p->noise_floor));
+ if (s->measure_perchannel & MEASURE_NOISE_FLOOR_COUNT)
+ set_meta(metadata, c + 1, "Noise_floor_count", "%f", p->noise_floor_count);
if (s->measure_perchannel & MEASURE_BIT_DEPTH) {
bit_depth(s, p->mask, p->imask, &depth);
set_meta(metadata, c + 1, "Bit_depth", "%f", depth.num);
set_meta(metadata, 0, "Overall.Flat_factor", "%f", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
if (s->measure_overall & MEASURE_PEAK_COUNT)
set_meta(metadata, 0, "Overall.Peak_count", "%f", (float)(min_count + max_count) / (double)s->nb_channels);
+ if (s->measure_overall & MEASURE_NOISE_FLOOR)
+ set_meta(metadata, 0, "Overall.Noise_floor", "%f", LINEAR_TO_DB(noise_floor));
+ if (s->measure_overall & MEASURE_NOISE_FLOOR_COUNT)
+ set_meta(metadata, 0, "Overall.Noise_floor_count", "%f", noise_floor_count / (double)s->nb_channels);
if (s->measure_overall & MEASURE_BIT_DEPTH) {
bit_depth(s, mask, imask, &depth);
set_meta(metadata, 0, "Overall.Bit_depth", "%f", depth.num);
}
#define UPDATE_STATS_P(type, update_func, update_float, channel_func) \
- for (int c = 0; c < channels; c++) { \
+ for (int c = start; c < end; c++) { \
ChannelStats *p = &s->chstats[c]; \
const type *src = (const type *)data[c]; \
const type * const srcend = src + samples; \
}
#define UPDATE_STATS_I(type, update_func, update_float, channel_func) \
- for (int c = 0; c < channels; c++) { \
+ for (int c = start; c < end; c++) { \
ChannelStats *p = &s->chstats[c]; \
const type *src = (const type *)data[0]; \
const type * const srcend = src + samples * channels; \
UPDATE_STATS_##planar(type, update_minmax(s, p, sample), , p->nmin = p->min normalizer_suffix; p->nmax = p->max normalizer_suffix;); \
}
-static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
+static int filter_channel(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
{
- AudioStatsContext *s = inlink->dst->priv;
- AVDictionary **metadata = &buf->metadata;
+ AudioStatsContext *s = ctx->priv;
+ AVFilterLink *inlink = ctx->inputs[0];
+ AVFrame *buf = arg;
+ const uint8_t * const * const data = (const uint8_t * const *)buf->extended_data;
const int channels = s->nb_channels;
const int samples = buf->nb_samples;
- const uint8_t * const * const data = (const uint8_t * const *)buf->extended_data;
-
- if (s->reset_count > 0) {
- if (s->nb_frames >= s->reset_count) {
- reset_stats(s);
- s->nb_frames = 0;
- }
- s->nb_frames++;
- }
+ const int start = (buf->channels * jobnr) / nb_jobs;
+ const int end = (buf->channels * (jobnr+1)) / nb_jobs;
switch (inlink->format) {
case AV_SAMPLE_FMT_DBLP:
break;
}
+ return 0;
+}
+
+static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
+{
+ AVFilterContext *ctx = inlink->dst;
+ AudioStatsContext *s = ctx->priv;
+ AVDictionary **metadata = &buf->metadata;
+
+ if (s->reset_count > 0) {
+ if (s->nb_frames >= s->reset_count) {
+ reset_stats(s);
+ s->nb_frames = 0;
+ }
+ s->nb_frames++;
+ }
+
+ ctx->internal->execute(ctx, filter_channel, buf, NULL, FFMIN(inlink->channels, ff_filter_get_nb_threads(ctx)));
+
if (s->metadata)
set_metadata(s, metadata);
static void print_stats(AVFilterContext *ctx)
{
AudioStatsContext *s = ctx->priv;
- uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0;
+ uint64_t mask = 0, imask = 0xFFFFFFFFFFFFFFFF, min_count = 0, max_count = 0, nb_samples = 0, noise_floor_count = 0;
uint64_t nb_nans = 0, nb_infs = 0, nb_denormals = 0;
double min_runs = 0, max_runs = 0,
min = DBL_MAX, max =-DBL_MAX, min_diff = DBL_MAX, max_diff = 0,
diff1_sum = 0,
sigma_x = 0,
sigma_x2 = 0,
+ noise_floor = 0,
min_sigma_x2 = DBL_MAX,
max_sigma_x2 =-DBL_MAX;
AVRational depth;
max_sigma_x2 = FFMAX(max_sigma_x2, p->max_sigma_x2);
sigma_x += p->sigma_x;
sigma_x2 += p->sigma_x2;
+ noise_floor = FFMAX(noise_floor, p->noise_floor);
min_count += p->min_count;
max_count += p->max_count;
+ noise_floor_count += p->noise_floor_count;
min_runs += p->min_runs;
max_runs += p->max_runs;
mask |= p->mask;
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)));
if (s->measure_perchannel & MEASURE_PEAK_COUNT)
av_log(ctx, AV_LOG_INFO, "Peak count: %"PRId64"\n", p->min_count + p->max_count);
+ if (s->measure_perchannel & MEASURE_NOISE_FLOOR)
+ av_log(ctx, AV_LOG_INFO, "Noise floor dB: %f\n", LINEAR_TO_DB(p->noise_floor));
+ if (s->measure_perchannel & MEASURE_NOISE_FLOOR_COUNT)
+ av_log(ctx, AV_LOG_INFO, "Noise floor count: %"PRId64"\n", p->noise_floor_count);
if (s->measure_perchannel & MEASURE_BIT_DEPTH) {
bit_depth(s, p->mask, p->imask, &depth);
av_log(ctx, AV_LOG_INFO, "Bit depth: %u/%u\n", depth.num, depth.den);
av_log(ctx, AV_LOG_INFO, "Flat factor: %f\n", LINEAR_TO_DB((min_runs + max_runs) / (min_count + max_count)));
if (s->measure_overall & MEASURE_PEAK_COUNT)
av_log(ctx, AV_LOG_INFO, "Peak count: %f\n", (min_count + max_count) / (double)s->nb_channels);
+ if (s->measure_overall & MEASURE_NOISE_FLOOR)
+ av_log(ctx, AV_LOG_INFO, "Noise floor dB: %f\n", LINEAR_TO_DB(noise_floor));
+ if (s->measure_overall & MEASURE_NOISE_FLOOR_COUNT)
+ av_log(ctx, AV_LOG_INFO, "Noise floor count: %f\n", noise_floor_count / (double)s->nb_channels);
if (s->measure_overall & MEASURE_BIT_DEPTH) {
bit_depth(s, mask, imask, &depth);
av_log(ctx, AV_LOG_INFO, "Bit depth: %u/%u\n", depth.num, depth.den);
if (s->nb_channels)
print_stats(ctx);
+ if (s->chstats) {
+ for (int i = 0; i < s->nb_channels; i++) {
+ ChannelStats *p = &s->chstats[i];
+
+ av_freep(&p->win_samples);
+ }
+ }
av_freep(&s->chstats);
}
{ NULL }
};
-AVFilter ff_af_astats = {
+const AVFilter ff_af_astats = {
.name = "astats",
.description = NULL_IF_CONFIG_SMALL("Show time domain statistics about audio frames."),
.query_formats = query_formats,
.uninit = uninit,
.inputs = astats_inputs,
.outputs = astats_outputs,
+ .flags = AVFILTER_FLAG_SLICE_THREADS,
};