Go to the documentation of this file.
32 #define MEASURE_ALL UINT_MAX
33 #define MEASURE_NONE 0
34 #define MEASURE_MEAN (1 << 0)
35 #define MEASURE_VARIANCE (1 << 1)
36 #define MEASURE_CENTROID (1 << 2)
37 #define MEASURE_SPREAD (1 << 3)
38 #define MEASURE_SKEWNESS (1 << 4)
39 #define MEASURE_KURTOSIS (1 << 5)
40 #define MEASURE_ENTROPY (1 << 6)
41 #define MEASURE_FLATNESS (1 << 7)
42 #define MEASURE_CREST (1 << 8)
43 #define MEASURE_FLUX (1 << 9)
44 #define MEASURE_SLOPE (1 << 10)
45 #define MEASURE_DECREASE (1 << 11)
46 #define MEASURE_ROLLOFF (1 << 12)
83 #define OFFSET(x) offsetof(AudioSpectralStatsContext, x)
84 #define A AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
114 float overlap,
scale = 1.f;
119 sizeof(*
s->window_func_lut));
120 if (!
s->window_func_lut)
123 if (
s->overlap == 1.f)
124 s->overlap = overlap;
126 s->hop_size =
s->win_size * (1.f -
s->overlap);
127 if (
s->hop_size <= 0)
130 s->stats =
av_calloc(
s->nb_channels,
sizeof(*
s->stats));
138 s->magnitude =
av_calloc(
s->nb_channels,
sizeof(*
s->magnitude));
142 s->prev_magnitude =
av_calloc(
s->nb_channels,
sizeof(*
s->prev_magnitude));
143 if (!
s->prev_magnitude)
146 s->fft_in =
av_calloc(
s->nb_channels,
sizeof(*
s->fft_in));
150 s->fft_out =
av_calloc(
s->nb_channels,
sizeof(*
s->fft_out));
154 for (
int ch = 0; ch <
s->nb_channels; ch++) {
159 s->fft_in[ch] =
av_calloc(
s->win_size,
sizeof(**
s->fft_in));
163 s->fft_out[ch] =
av_calloc(
s->win_size,
sizeof(**
s->fft_out));
167 s->magnitude[ch] =
av_calloc(
s->win_size,
sizeof(**
s->magnitude));
168 if (!
s->magnitude[ch])
171 s->prev_magnitude[ch] =
av_calloc(
s->win_size,
sizeof(**
s->prev_magnitude));
172 if (!
s->prev_magnitude[ch])
184 const char *fmt,
float val)
191 snprintf(key2,
sizeof(key2),
"lavfi.aspectralstats.%d.%s", chan,
key);
193 snprintf(key2,
sizeof(key2),
"lavfi.aspectralstats.%s",
key);
199 for (
int ch = 0; ch <
s->nb_channels; ch++) {
205 set_meta(metadata, ch + 1,
"variance",
"%g",
stats->variance);
207 set_meta(metadata, ch + 1,
"centroid",
"%g",
stats->centroid);
211 set_meta(metadata, ch + 1,
"skewness",
"%g",
stats->skewness);
213 set_meta(metadata, ch + 1,
"kurtosis",
"%g",
stats->kurtosis);
217 set_meta(metadata, ch + 1,
"flatness",
"%g",
stats->flatness);
225 set_meta(metadata, ch + 1,
"decrease",
"%g",
stats->decrease);
235 for (
int n = 0; n <
size; n++)
250 for (
int n = 0; n <
size; n++)
259 float num = 0.f, den = 0.f;
261 for (
int n = 0; n <
size; n++) {
262 num += spectral[n] * n *
scale;
266 if (den <= FLT_EPSILON)
274 float num = 0.f, den = 0.f;
276 for (
int n = 0; n <
size; n++) {
277 num += spectral[n] *
sqrf(n *
scale - centroid);
281 if (den <= FLT_EPSILON)
283 return sqrtf(num / den);
294 float num = 0.f, den = 0.f;
296 for (
int n = 0; n <
size; n++) {
297 num += spectral[n] *
cbrf(n *
scale - centroid);
302 if (den <= FLT_EPSILON)
310 float num = 0.f, den = 0.f;
312 for (
int n = 0; n <
size; n++) {
318 if (den <= FLT_EPSILON)
325 float num = 0.f, den = 0.f;
327 for (
int n = 0; n <
size; n++) {
328 num += spectral[n] * logf(spectral[n] + FLT_EPSILON);
332 if (den <= FLT_EPSILON)
339 float num = 0.f, den = 0.f;
341 for (
int n = 0; n <
size; n++) {
342 float v = FLT_EPSILON + spectral[n];
350 if (den <= FLT_EPSILON)
359 for (
int n = 0; n <
size; n++) {
365 if (
mean <= FLT_EPSILON)
370 static float spectral_flux(
const float *
const spectral,
const float *
const prev_spectral,
371 int size,
int max_freq)
375 for (
int n = 0; n <
size; n++)
376 sum +=
sqrf(spectral[n] - prev_spectral[n]);
383 const float mean_freq =
size * 0.5f;
384 float mean_spectral = 0.f, num = 0.f, den = 0.f;
386 for (
int n = 0; n <
size; n++)
387 mean_spectral += spectral[n];
388 mean_spectral /=
size;
390 for (
int n = 0; n <
size; n++) {
391 num += ((n - mean_freq) / mean_freq) * (spectral[n] - mean_spectral);
392 den +=
sqrf((n - mean_freq) / mean_freq);
395 if (
fabsf(den) <= FLT_EPSILON)
402 float num = 0.f, den = 0.f;
404 for (
int n = 1; n <
size; n++) {
405 num += (spectral[n] - spectral[0]) / n;
409 if (den <= FLT_EPSILON)
417 float norm = 0.f, sum = 0.f;
420 for (
int n = 0; n <
size; n++)
424 for (
int n = 0; n <
size; n++) {
438 const float *window_func_lut =
s->window_func_lut;
441 const int start = (
channels * jobnr) / nb_jobs;
442 const int end = (
channels * (jobnr+1)) / nb_jobs;
443 const int offset =
s->win_size -
s->hop_size;
445 for (
int ch = start; ch < end; ch++) {
446 float *
window = (
float *)
s->window->extended_data[ch];
450 float *magnitude =
s->magnitude[ch];
451 float *prev_magnitude =
s->prev_magnitude[ch];
452 const float scale = 1.f /
s->win_size;
458 for (
int n = 0; n <
s->win_size; n++) {
459 fft_in[n].re =
window[n] * window_func_lut[n];
463 s->tx_fn(
s->fft[ch], fft_out, fft_in,
sizeof(*fft_in));
465 for (
int n = 0; n <
s->win_size / 2; n++) {
466 fft_out[n].re *=
scale;
467 fft_out[n].im *=
scale;
470 for (
int n = 0; n <
s->win_size / 2; n++)
471 magnitude[n] = hypotf(fft_out[n].
re, fft_out[n].
im);
500 memcpy(prev_magnitude, magnitude,
s->win_size *
sizeof(
float));
531 metadata = &
out->metadata;
579 for (
int ch = 0; ch <
s->nb_channels; ch++) {
588 if (
s->prev_magnitude)
619 .
name =
"aspectralstats",
622 .priv_class = &aspectralstats_class,
AVFrame * ff_get_audio_buffer(AVFilterLink *link, int nb_samples)
Request an audio samples buffer with a specific set of permissions.
@ AV_SAMPLE_FMT_FLTP
float, planar
static float spectral_mean(const float *const spectral, int size, int max_freq)
Filter the word “frame” indicates either a video frame or a group of audio as stored in an AVFrame structure Format for each input and each output the list of supported formats For video that means pixel format For audio that means channel sample they are references to shared objects When the negotiation mechanism computes the intersection of the formats supported at each end of a all references to both lists are replaced with a reference to the intersection And when a single format is eventually chosen for a link amongst the remaining all references to the list are updated That means that if a filter requires that its input and output have the same format amongst a supported all it has to do is use a reference to the same list of formats query_formats can leave some formats unset and return AVERROR(EAGAIN) to cause the negotiation mechanism toagain later. That can be used by filters with complex requirements to use the format negotiated on one link to set the formats supported on another. Frame references ownership and permissions
static void stats(AVPacket *const *in, int n_in, unsigned *_max, unsigned *_sum)
int ff_filter_frame(AVFilterLink *link, AVFrame *frame)
Send a frame of data to the next filter.
#define FILTER_SINGLE_SAMPLEFMT(sample_fmt_)
The exact code depends on how similar the blocks are and how related they are to the and needs to apply these operations to the correct inlink or outlink if there are several Macros are available to factor that when no extra processing is inlink
void av_frame_free(AVFrame **frame)
Free the frame and any dynamically allocated objects in it, e.g.
This structure describes decoded (raw) audio or video data.
#define WIN_FUNC_OPTION(win_func_opt_name, win_func_offset, flag, default_window_func)
const char * name
Filter name.
int nb_channels
Number of channels in this layout.
static const AVFilterPad aspectralstats_outputs[]
A link between two filters.
#define FF_FILTER_FORWARD_STATUS_BACK(outlink, inlink)
Forward the status on an output link to an input link.
av_cold int av_tx_init(AVTXContext **ctx, av_tx_fn *tx, enum AVTXType type, int inv, int len, const void *scale, uint64_t flags)
Initialize a transform context with the given configuration (i)MDCTs with an odd length are currently...
static float spectral_entropy(const float *const spectral, int size, int max_freq)
AVComplexFloat ** fft_out
static SDL_Window * window
void * priv
private data for use by the filter
static float spectral_variance(const float *const spectral, int size, int max_freq, float mean)
static double val(void *priv, double ch)
static av_always_inline float scale(float x, float s)
static __device__ float fabsf(float a)
A filter pad used for either input or output.
const AVFilter ff_af_aspectralstats
static float spectral_crest(const float *const spectral, int size, int max_freq)
static int filter_channel(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
void(* av_tx_fn)(AVTXContext *s, void *out, void *in, ptrdiff_t stride)
Function pointer to a function to perform the transform.
static float spectral_flatness(const float *const spectral, int size, int max_freq)
@ AV_TX_FLOAT_FFT
Standard complex to complex FFT with sample data type of AVComplexFloat, AVComplexDouble or AVComplex...
#define FILTER_INPUTS(array)
#define av_realloc_f(p, o, n)
static float spectral_slope(const float *const spectral, int size, int max_freq)
Describe the class of an AVClass context structure.
int ff_inlink_consume_samples(AVFilterLink *link, unsigned min, unsigned max, AVFrame **rframe)
Take samples from the link's FIFO and update the link's stats.
int av_frame_copy_props(AVFrame *dst, const AVFrame *src)
Copy only "metadata" fields from src to dst.
static __device__ float sqrtf(float a)
static void generate_window_func(float *lut, int N, int win_func, float *overlap)
static const AVOption aspectralstats_options[]
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification.
int av_frame_copy(AVFrame *dst, const AVFrame *src)
Copy the frame data from src to dst.
int sample_rate
Sample rate of the audio data.
float fmaxf(float, float)
static int filter_frame(AVFilterLink *inlink, AVFrame *in)
int av_frame_is_writable(AVFrame *frame)
Check if the frame data is writable.
AVFilterContext * src
source filter
The reader does not expect b to be semantically here and if the code is changed by maybe adding a a division or other the signedness will almost certainly be mistaken To avoid this confusion a new type was SUINT is the C unsigned type but it holds a signed int to use the same example SUINT a
static float spectral_decrease(const float *const spectral, int size, int max_freq)
it s the only field you need to keep assuming you have a context There is some magic you don t need to care about around this just let it vf offset
FF_FILTER_FORWARD_WANTED(outlink, inlink)
static int activate(AVFilterContext *ctx)
static float spectral_centroid(const float *const spectral, int size, int max_freq)
av_cold void av_tx_uninit(AVTXContext **ctx)
Frees a context and sets *ctx to NULL, does nothing when *ctx == NULL.
static av_cold void uninit(AVFilterContext *ctx)
static float spectral_rolloff(const float *const spectral, int size, int max_freq)
int nb_samples
number of audio samples (per channel) described by this frame
uint8_t ** extended_data
pointers to the data planes/channels.
static float cbrf(float a)
static float spectral_flux(const float *const spectral, const float *const prev_spectral, int size, int max_freq)
int ff_filter_get_nb_threads(AVFilterContext *ctx)
Get number of threads for current filter instance.
it s the only field you need to keep assuming you have a context There is some magic you don t need to care about around this just let it vf default value
const char * name
Pad name.
int ff_inlink_queued_samples(AVFilterLink *link)
void * av_calloc(size_t nmemb, size_t size)
static int config_output(AVFilterLink *outlink)
static const AVFilterPad aspectralstats_inputs[]
static void set_meta(AVDictionary **metadata, int chan, const char *key, const char *fmt, float val)
static float spectral_spread(const float *const spectral, int size, int max_freq, float centroid)
static float mean(const float *input, int size)
static float spectral_skewness(const float *const spectral, int size, int max_freq, float centroid, float spread)
#define AVFILTER_FLAG_SLICE_THREADS
The filter supports multithreading by splitting frames into multiple parts and processing them concur...
AVChannelLayout ch_layout
channel layout of current buffer (see libavutil/channel_layout.h)
FF_FILTER_FORWARD_STATUS(inlink, outlink)
#define FILTER_OUTPUTS(array)
int av_dict_set(AVDictionary **pm, const char *key, const char *value, int flags)
Set the given entry in *pm, overwriting an existing entry.
static void set_metadata(AudioSpectralStatsContext *s, AVDictionary **metadata)
static float sqrf(float a)
ChannelSpectralStats * stats
AVFILTER_DEFINE_CLASS(aspectralstats)
static av_always_inline int ff_filter_execute(AVFilterContext *ctx, avfilter_action_func *func, void *arg, int *ret, int nb_jobs)
static float spectral_kurtosis(const float *const spectral, int size, int max_freq, float centroid, float spread)
void ff_filter_set_ready(AVFilterContext *filter, unsigned priority)
Mark a filter ready and schedule it for activation.