avfilter/af_silenceremove: add support for more sample formats

pull/360/head
Paul B Mahol 3 years ago
parent 6f0f7bc7fe
commit d08deb11d2
  1. 461
      libavfilter/af_silenceremove.c

@ -69,16 +69,16 @@ typedef struct SilenceRemoveContext {
int64_t stop_silence_opt;
int stop_mode;
double *start_holdoff;
double *start_silence_hold;
AVFrame *start_holdoff;
AVFrame *start_silence_hold;
size_t start_holdoff_offset;
size_t start_holdoff_end;
size_t start_silence_offset;
size_t start_silence_end;
int start_found_periods;
double *stop_holdoff;
double *stop_silence_hold;
AVFrame *stop_holdoff;
AVFrame *stop_silence_hold;
size_t stop_holdoff_offset;
size_t stop_holdoff_end;
size_t stop_silence_offset;
@ -96,8 +96,10 @@ typedef struct SilenceRemoveContext {
int64_t next_pts;
int detection;
void (*update)(struct SilenceRemoveContext *s, double sample);
double(*compute)(struct SilenceRemoveContext *s, double sample);
void (*update)(struct SilenceRemoveContext *s, AVFrame *frame, int ch, int offset);
double (*compute)(struct SilenceRemoveContext *s, AVFrame *frame, int ch, int offset);
void (*copy)(struct SilenceRemoveContext *s, AVFrame *out, AVFrame *in,
int ch, int out_offset, int in_offset);
} SilenceRemoveContext;
#define OFFSET(x) offsetof(SilenceRemoveContext, x)
@ -125,8 +127,158 @@ static const AVOption silenceremove_options[] = {
AVFILTER_DEFINE_CLASS(silenceremove);
static double compute_peak(SilenceRemoveContext *s, double sample)
static void copy_double(SilenceRemoveContext *s, AVFrame *out, AVFrame *in,
int ch, int out_offset, int in_offset)
{
const double *srcp = (const double *)in->data[0];
const double src = srcp[in->channels * in_offset + ch];
double *dstp = (double *)out->data[0];
dstp[out->channels * out_offset + ch] = src;
}
static void copy_doublep(SilenceRemoveContext *s, AVFrame *out, AVFrame *in,
int ch, int out_offset, int in_offset)
{
const double *srcp = (const double *)in->data[ch];
const double src = srcp[in_offset];
double *dstp = (double *)out->data[ch];
dstp[out_offset] = src;
}
static void copy_float(SilenceRemoveContext *s, AVFrame *out, AVFrame *in,
int ch, int out_offset, int in_offset)
{
const float *srcp = (const float *)in->data[0];
const float src = srcp[in->channels * in_offset + ch];
float *dstp = (float *)out->data[0];
dstp[out->channels * out_offset + ch] = src;
}
static void copy_floatp(SilenceRemoveContext *s, AVFrame *out, AVFrame *in,
int ch, int out_offset, int in_offset)
{
const float *srcp = (const float *)in->data[ch];
const float src = srcp[in_offset];
float *dstp = (float *)out->data[ch];
dstp[out_offset] = src;
}
static double compute_peak_double(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const double *samples = (const double *)frame->data[0];
double sample = samples[frame->channels * offset + ch];
double new_sum;
new_sum = s->sum;
new_sum -= *s->window_current;
new_sum += fabs(sample);
return new_sum / s->window_size;
}
static void update_peak_double(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const double *samples = (const double *)frame->data[0];
double sample = samples[frame->channels * offset + ch];
s->sum -= *s->window_current;
*s->window_current = fabs(sample);
s->sum += *s->window_current;
s->window_current++;
if (s->window_current >= s->window_end)
s->window_current = s->window;
}
static double compute_peak_float(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const float *samples = (const float *)frame->data[0];
float sample = samples[frame->channels * offset + ch];
float new_sum;
new_sum = s->sum;
new_sum -= *s->window_current;
new_sum += fabs(sample);
return new_sum / s->window_size;
}
static void update_peak_float(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const float *samples = (const float *)frame->data[0];
float sample = samples[frame->channels * offset + ch];
s->sum -= *s->window_current;
*s->window_current = fabs(sample);
s->sum += *s->window_current;
s->window_current++;
if (s->window_current >= s->window_end)
s->window_current = s->window;
}
static double compute_rms_double(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const double *samples = (const double *)frame->data[0];
double sample = samples[frame->channels * offset + ch];
double new_sum;
new_sum = s->sum;
new_sum -= *s->window_current;
new_sum += sample * sample;
return sqrt(new_sum / s->window_size);
}
static void update_rms_double(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const double *samples = (const double *)frame->data[0];
double sample = samples[frame->channels * offset + ch];
s->sum -= *s->window_current;
*s->window_current = sample * sample;
s->sum += *s->window_current;
s->window_current++;
if (s->window_current >= s->window_end)
s->window_current = s->window;
}
static double compute_rms_float(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const float *samples = (const float *)frame->data[0];
float sample = samples[frame->channels * offset + ch];
float new_sum;
new_sum = s->sum;
new_sum -= *s->window_current;
new_sum += sample * sample;
return sqrtf(new_sum / s->window_size);
}
static void update_rms_float(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const float *samples = (const float *)frame->data[0];
float sample = samples[frame->channels * offset + ch];
s->sum -= *s->window_current;
*s->window_current = sample * sample;
s->sum += *s->window_current;
s->window_current++;
if (s->window_current >= s->window_end)
s->window_current = s->window;
}
static double compute_peak_doublep(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const double *samples = (const double *)frame->data[ch];
double sample = samples[offset];
double new_sum;
new_sum = s->sum;
@ -136,8 +288,38 @@ static double compute_peak(SilenceRemoveContext *s, double sample)
return new_sum / s->window_size;
}
static void update_peak(SilenceRemoveContext *s, double sample)
static void update_peak_doublep(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const double *samples = (const double *)frame->data[ch];
double sample = samples[offset];
s->sum -= *s->window_current;
*s->window_current = fabs(sample);
s->sum += *s->window_current;
s->window_current++;
if (s->window_current >= s->window_end)
s->window_current = s->window;
}
static double compute_peak_floatp(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const float *samples = (const float *)frame->data[ch];
float sample = samples[offset];
float new_sum;
new_sum = s->sum;
new_sum -= *s->window_current;
new_sum += fabs(sample);
return new_sum / s->window_size;
}
static void update_peak_floatp(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const float *samples = (const float *)frame->data[ch];
float sample = samples[offset];
s->sum -= *s->window_current;
*s->window_current = fabs(sample);
s->sum += *s->window_current;
@ -147,8 +329,10 @@ static void update_peak(SilenceRemoveContext *s, double sample)
s->window_current = s->window;
}
static double compute_rms(SilenceRemoveContext *s, double sample)
static double compute_rms_doublep(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const double *samples = (const double *)frame->data[ch];
double sample = samples[offset];
double new_sum;
new_sum = s->sum;
@ -158,8 +342,38 @@ static double compute_rms(SilenceRemoveContext *s, double sample)
return sqrt(new_sum / s->window_size);
}
static void update_rms(SilenceRemoveContext *s, double sample)
static void update_rms_doublep(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const double *samples = (const double *)frame->data[ch];
double sample = samples[offset];
s->sum -= *s->window_current;
*s->window_current = sample * sample;
s->sum += *s->window_current;
s->window_current++;
if (s->window_current >= s->window_end)
s->window_current = s->window;
}
static double compute_rms_floatp(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const float *samples = (const float *)frame->data[ch];
float sample = samples[offset];
float new_sum;
new_sum = s->sum;
new_sum -= *s->window_current;
new_sum += sample * sample;
return sqrtf(new_sum / s->window_size);
}
static void update_rms_floatp(SilenceRemoveContext *s, AVFrame *frame, int ch, int offset)
{
const float *samples = (const float *)frame->data[ch];
float sample = samples[offset];
s->sum -= *s->window_current;
*s->window_current = sample * sample;
s->sum += *s->window_current;
@ -178,17 +392,6 @@ static av_cold int init(AVFilterContext *ctx)
s->restart = 1;
}
switch (s->detection) {
case D_PEAK:
s->update = update_peak;
s->compute = compute_peak;
break;
case D_RMS:
s->update = update_rms;
s->compute = compute_rms;
break;
}
return 0;
}
@ -223,15 +426,13 @@ static int config_input(AVFilterLink *inlink)
s->stop_silence = av_rescale(s->stop_silence_opt, inlink->sample_rate,
AV_TIME_BASE);
s->start_holdoff = av_malloc_array(FFMAX(s->start_duration, 1),
sizeof(*s->start_holdoff) *
inlink->channels);
s->start_holdoff = ff_get_audio_buffer(ctx->outputs[0],
FFMAX(s->start_duration, 1));
if (!s->start_holdoff)
return AVERROR(ENOMEM);
s->start_silence_hold = av_malloc_array(FFMAX(s->start_silence, 1),
sizeof(*s->start_silence_hold) *
inlink->channels);
s->start_silence_hold = ff_get_audio_buffer(ctx->outputs[0],
FFMAX(s->start_silence, 1));
if (!s->start_silence_hold)
return AVERROR(ENOMEM);
@ -239,15 +440,13 @@ static int config_input(AVFilterLink *inlink)
s->start_holdoff_end = 0;
s->start_found_periods = 0;
s->stop_holdoff = av_malloc_array(FFMAX(s->stop_duration, 1),
sizeof(*s->stop_holdoff) *
inlink->channels);
s->stop_holdoff = ff_get_audio_buffer(ctx->outputs[0],
FFMAX(s->stop_duration, 1));
if (!s->stop_holdoff)
return AVERROR(ENOMEM);
s->stop_silence_hold = av_malloc_array(FFMAX(s->stop_silence, 1),
sizeof(*s->stop_silence_hold) *
inlink->channels);
s->stop_silence_hold = ff_get_audio_buffer(ctx->outputs[0],
FFMAX(s->stop_silence, 1));
if (!s->stop_silence_hold)
return AVERROR(ENOMEM);
@ -260,6 +459,61 @@ static int config_input(AVFilterLink *inlink)
else
s->mode = SILENCE_COPY;
switch (inlink->format) {
case AV_SAMPLE_FMT_DBL:
s->copy = copy_double;
switch (s->detection) {
case D_PEAK:
s->update = update_peak_double;
s->compute = compute_peak_double;
break;
case D_RMS:
s->update = update_rms_double;
s->compute = compute_rms_double;
break;
}
break;
case AV_SAMPLE_FMT_FLT:
s->copy = copy_float;
switch (s->detection) {
case D_PEAK:
s->update = update_peak_float;
s->compute = compute_peak_float;
break;
case D_RMS:
s->update = update_rms_float;
s->compute = compute_rms_float;
break;
}
break;
case AV_SAMPLE_FMT_DBLP:
s->copy = copy_doublep;
switch (s->detection) {
case D_PEAK:
s->update = update_peak_doublep;
s->compute = compute_peak_doublep;
break;
case D_RMS:
s->update = update_rms_doublep;
s->compute = compute_rms_doublep;
break;
}
break;
case AV_SAMPLE_FMT_FLTP:
s->copy = copy_floatp;
switch (s->detection) {
case D_PEAK:
s->update = update_peak_floatp;
s->compute = compute_peak_floatp;
break;
case D_RMS:
s->update = update_rms_floatp;
s->compute = compute_rms_floatp;
break;
}
break;
}
return 0;
}
@ -270,7 +524,7 @@ static void flush(SilenceRemoveContext *s,
AVFrame *silence;
if (*nb_samples_written) {
out->nb_samples = *nb_samples_written / outlink->channels;
out->nb_samples = *nb_samples_written;
out->pts = s->next_pts;
s->next_pts += av_rescale_q(out->nb_samples,
@ -288,22 +542,24 @@ static void flush(SilenceRemoveContext *s,
if (s->stop_silence_end <= 0 || !flush_silence)
return;
silence = ff_get_audio_buffer(outlink, s->stop_silence_end / outlink->channels);
silence = ff_get_audio_buffer(outlink, s->stop_silence_end);
if (!silence) {
*ret = AVERROR(ENOMEM);
return;
}
if (s->stop_silence_offset < s->stop_silence_end) {
memcpy(silence->data[0],
&s->stop_silence_hold[s->stop_silence_offset],
(s->stop_silence_end - s->stop_silence_offset) * sizeof(double));
av_samples_copy(silence->data, s->stop_silence_hold->data, 0,
s->stop_silence_offset,
s->stop_silence_end - s->stop_silence_offset,
outlink->channels, outlink->format);
}
if (s->stop_silence_offset > 0) {
memcpy(silence->data[0] + (s->stop_silence_end - s->stop_silence_offset) * sizeof(double),
&s->stop_silence_hold[0],
s->stop_silence_offset * sizeof(double));
av_samples_copy(silence->data, s->stop_silence_hold->data,
s->stop_silence_end - s->stop_silence_offset,
0, s->stop_silence_offset,
outlink->channels, outlink->format);
}
s->stop_silence_offset = 0;
@ -324,7 +580,6 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
SilenceRemoveContext *s = ctx->priv;
int i, j, threshold, ret = 0;
int nbs, nb_samples_read, nb_samples_written;
double *obuf, *ibuf = (double *)in->data[0];
AVFrame *out;
nb_samples_read = nb_samples_written = 0;
@ -335,7 +590,7 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
switch (s->mode) {
case SILENCE_TRIM:
silence_trim:
nbs = in->nb_samples - nb_samples_read / outlink->channels;
nbs = in->nb_samples - nb_samples_read;
if (!nbs)
break;
@ -343,23 +598,24 @@ silence_trim:
if (s->start_mode == T_ANY) {
threshold = 0;
for (j = 0; j < outlink->channels; j++) {
threshold |= s->compute(s, ibuf[j]) > s->start_threshold;
threshold |= s->compute(s, in, j, nb_samples_read) > s->start_threshold;
}
} else {
threshold = 1;
for (j = 0; j < outlink->channels; j++) {
threshold &= s->compute(s, ibuf[j]) > s->start_threshold;
threshold &= s->compute(s, in, j, nb_samples_read) > s->start_threshold;
}
}
if (threshold) {
for (j = 0; j < outlink->channels; j++) {
s->update(s, *ibuf);
s->start_holdoff[s->start_holdoff_end++] = *ibuf++;
s->update(s, in, j, nb_samples_read);
s->copy(s, s->start_holdoff, in, j, s->start_holdoff_end, nb_samples_read);
}
nb_samples_read += outlink->channels;
s->start_holdoff_end++;
nb_samples_read++;
if (s->start_holdoff_end >= s->start_duration * outlink->channels) {
if (s->start_holdoff_end >= s->start_duration) {
if (++s->start_found_periods >= s->start_periods) {
s->mode = SILENCE_TRIM_FLUSH;
goto silence_trim_flush;
@ -374,30 +630,30 @@ silence_trim:
s->start_holdoff_end = 0;
for (j = 0; j < outlink->channels; j++) {
s->update(s, ibuf[j]);
s->update(s, in, j, nb_samples_read);
if (s->start_silence)
s->copy(s, s->start_silence_hold, in, j, s->start_silence_offset, nb_samples_read);
}
nb_samples_read++;
s->start_silence_offset++;
if (s->start_silence) {
s->start_silence_hold[s->start_silence_offset++] = ibuf[j];
s->start_silence_end = FFMIN(s->start_silence_end + 1, outlink->channels * s->start_silence);
if (s->start_silence_offset >= outlink->channels * s->start_silence) {
s->start_silence_end = FFMIN(s->start_silence_end + 1, s->start_silence);
if (s->start_silence_offset >= s->start_silence)
s->start_silence_offset = 0;
}
}
}
ibuf += outlink->channels;
nb_samples_read += outlink->channels;
}
}
break;
case SILENCE_TRIM_FLUSH:
silence_trim_flush:
nbs = s->start_holdoff_end - s->start_holdoff_offset;
nbs -= nbs % outlink->channels;
if (!nbs)
break;
out = ff_get_audio_buffer(outlink, nbs / outlink->channels + s->start_silence_end / outlink->channels);
out = ff_get_audio_buffer(outlink, nbs + s->start_silence_end);
if (!out) {
av_frame_free(&in);
return AVERROR(ENOMEM);
@ -405,21 +661,24 @@ silence_trim_flush:
if (s->start_silence_end > 0) {
if (s->start_silence_offset < s->start_silence_end) {
memcpy(out->data[0],
&s->start_silence_hold[s->start_silence_offset],
(s->start_silence_end - s->start_silence_offset) * sizeof(double));
av_samples_copy(out->data, s->start_silence_hold->data, 0,
s->start_silence_offset,
s->start_silence_end - s->start_silence_offset,
outlink->channels, outlink->format);
}
if (s->start_silence_offset > 0) {
memcpy(out->data[0] + (s->start_silence_end - s->start_silence_offset) * sizeof(double),
&s->start_silence_hold[0],
s->start_silence_offset * sizeof(double));
av_samples_copy(out->data, s->start_silence_hold->data,
s->start_silence_end - s->start_silence_offset,
0, s->start_silence_offset,
outlink->channels, outlink->format);
}
}
memcpy(out->data[0] + s->start_silence_end * sizeof(double),
&s->start_holdoff[s->start_holdoff_offset],
nbs * sizeof(double));
av_samples_copy(out->data, s->start_holdoff->data,
s->start_silence_end,
s->start_holdoff_offset, nbs,
outlink->channels, outlink->format);
out->pts = s->next_pts;
s->next_pts += av_rescale_q(out->nb_samples,
@ -442,7 +701,7 @@ silence_trim_flush:
case SILENCE_COPY:
silence_copy:
nbs = in->nb_samples - nb_samples_read / outlink->channels;
nbs = in->nb_samples - nb_samples_read;
if (!nbs)
break;
@ -451,19 +710,18 @@ silence_copy:
av_frame_free(&in);
return AVERROR(ENOMEM);
}
obuf = (double *)out->data[0];
if (s->stop_periods) {
for (i = 0; i < nbs; i++) {
if (s->stop_mode == T_ANY) {
threshold = 0;
for (j = 0; j < outlink->channels; j++) {
threshold |= s->compute(s, ibuf[j]) > s->stop_threshold;
threshold |= s->compute(s, in, j, nb_samples_read) > s->stop_threshold;
}
} else {
threshold = 1;
for (j = 0; j < outlink->channels; j++) {
threshold &= s->compute(s, ibuf[j]) > s->stop_threshold;
threshold &= s->compute(s, in, j, nb_samples_read) > s->stop_threshold;
}
}
@ -473,27 +731,29 @@ silence_copy:
goto silence_copy_flush;
} else if (threshold) {
for (j = 0; j < outlink->channels; j++) {
s->update(s, *ibuf);
*obuf++ = *ibuf++;
s->update(s, in, j, nb_samples_read);
s->copy(s, out, in, j, nb_samples_written, nb_samples_read);
}
nb_samples_read += outlink->channels;
nb_samples_written += outlink->channels;
nb_samples_read++;
nb_samples_written++;
} else if (!threshold) {
for (j = 0; j < outlink->channels; j++) {
s->update(s, *ibuf);
s->update(s, in, j, nb_samples_read);
if (s->stop_silence) {
s->stop_silence_hold[s->stop_silence_offset++] = *ibuf;
s->stop_silence_end = FFMIN(s->stop_silence_end + 1, outlink->channels * s->stop_silence);
if (s->stop_silence_offset >= outlink->channels * s->stop_silence) {
s->copy(s, s->stop_silence_hold, in, j, s->stop_silence_offset, nb_samples_read);
s->stop_silence_end = FFMIN(s->stop_silence_end + 1, s->stop_silence);
if (s->stop_silence_offset >= s->stop_silence) {
s->stop_silence_offset = 0;
}
}
s->stop_holdoff[s->stop_holdoff_end++] = *ibuf++;
s->copy(s, s->stop_holdoff, in, j, s->stop_holdoff_end, nb_samples_read);
}
nb_samples_read += outlink->channels;
nb_samples_read++;
s->stop_holdoff_end++;
if (s->stop_holdoff_end >= s->stop_duration * outlink->channels) {
if (s->stop_holdoff_end >= s->stop_duration) {
if (++s->stop_found_periods >= s->stop_periods) {
s->stop_holdoff_offset = 0;
s->stop_holdoff_end = 0;
@ -523,7 +783,10 @@ silence_copy:
}
flush(s, out, outlink, &nb_samples_written, &ret, 0);
} else {
memcpy(obuf, ibuf, sizeof(double) * nbs * outlink->channels);
av_samples_copy(out->data, in->data,
nb_samples_written,
nb_samples_read, nbs,
outlink->channels, outlink->format);
out->pts = s->next_pts;
s->next_pts += av_rescale_q(out->nb_samples,
@ -537,18 +800,19 @@ silence_copy:
case SILENCE_COPY_FLUSH:
silence_copy_flush:
nbs = s->stop_holdoff_end - s->stop_holdoff_offset;
nbs -= nbs % outlink->channels;
if (!nbs)
break;
out = ff_get_audio_buffer(outlink, nbs / outlink->channels);
out = ff_get_audio_buffer(outlink, nbs);
if (!out) {
av_frame_free(&in);
return AVERROR(ENOMEM);
}
memcpy(out->data[0], &s->stop_holdoff[s->stop_holdoff_offset],
nbs * sizeof(double));
av_samples_copy(out->data, s->stop_holdoff->data, 0,
s->stop_holdoff_offset, nbs,
outlink->channels, outlink->format);
s->stop_holdoff_offset += nbs;
out->pts = s->next_pts;
@ -590,12 +854,13 @@ static int request_frame(AVFilterLink *outlink)
if (nbs) {
AVFrame *frame;
frame = ff_get_audio_buffer(outlink, nbs / outlink->channels);
frame = ff_get_audio_buffer(outlink, nbs);
if (!frame)
return AVERROR(ENOMEM);
memcpy(frame->data[0], &s->stop_holdoff[s->stop_holdoff_offset],
nbs * sizeof(double));
av_samples_copy(frame->data, s->stop_holdoff->data, 0,
s->stop_holdoff_offset, nbs,
outlink->channels, outlink->format);
frame->pts = s->next_pts;
s->next_pts += av_rescale_q(frame->nb_samples,
@ -614,7 +879,9 @@ static int query_formats(AVFilterContext *ctx)
AVFilterFormats *formats = NULL;
AVFilterChannelLayouts *layouts = NULL;
static const enum AVSampleFormat sample_fmts[] = {
AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_NONE
AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_DBLP,
AV_SAMPLE_FMT_NONE
};
int ret;
@ -642,10 +909,10 @@ static av_cold void uninit(AVFilterContext *ctx)
{
SilenceRemoveContext *s = ctx->priv;
av_freep(&s->start_holdoff);
av_freep(&s->start_silence_hold);
av_freep(&s->stop_holdoff);
av_freep(&s->stop_silence_hold);
av_frame_free(&s->start_holdoff);
av_frame_free(&s->start_silence_hold);
av_frame_free(&s->stop_holdoff);
av_frame_free(&s->stop_silence_hold);
av_freep(&s->window);
}

Loading…
Cancel
Save