virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2025-08-10 06:10:52 +02:00
Code Issues Releases Activity

avfilter/af_speechnorm: implement rms option

Browse Source
This commit is contained in:
Paul B Mahol
2022-11-06 14:03:27 +01:00
parent b48d2320f1
commit 7027101904
2 changed files with 31 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
doc/filters.texi
View File

@@ -6338,6 +6338,10 @@ option. When enabled any half-cycle of samples with their local peak value below
Link channels when calculating gain applied to each filtered channel sample, by default is disabled. Link channels when calculating gain applied to each filtered channel sample, by default is disabled.
When disabled each filtered channel gain calculation is independent, otherwise when this option When disabled each filtered channel gain calculation is independent, otherwise when this option
is enabled the minimum of all possible gains for each filtered channel is used. is enabled the minimum of all possible gains for each filtered channel is used.
@item rms, m
Set the expansion target RMS value. This specifies the highest allowed RMS level for the normalized
audio input. Default value is 0.0, thus disabled. Allowed range is from 0.0 to 1.0.
@end table @end table
@subsection Commands @subsection Commands

33
libavfilter/af_speechnorm.c
View File

@@ -46,6 +46,7 @@ typedef struct PeriodItem {
int size; int size;
int type; int type;
double max_peak; double max_peak;
double rms_sum;
} PeriodItem; } PeriodItem;
typedef struct ChannelContext { typedef struct ChannelContext {
@@ -54,6 +55,7 @@ typedef struct ChannelContext {
PeriodItem pi[MAX_ITEMS]; PeriodItem pi[MAX_ITEMS];
double gain_state; double gain_state;
double pi_max_peak; double pi_max_peak;
double pi_rms_sum;
int pi_start; int pi_start;
int pi_end; int pi_end;
int pi_size; int pi_size;
@@ -62,6 +64,7 @@ typedef struct ChannelContext {
typedef struct SpeechNormalizerContext { typedef struct SpeechNormalizerContext {
const AVClass *class; const AVClass *class;
double rms_value;
double peak_value; double peak_value;
double max_expansion; double max_expansion;
double max_compression; double max_compression;
@@ -110,6 +113,8 @@ static const AVOption speechnorm_options[] = {
{ "i", "set inverted filtering", OFFSET(invert), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS }, { "i", "set inverted filtering", OFFSET(invert), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS },
{ "link", "set linked channels filtering", OFFSET(link), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS }, { "link", "set linked channels filtering", OFFSET(link), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS },
{ "l", "set linked channels filtering", OFFSET(link), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS }, { "l", "set linked channels filtering", OFFSET(link), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS },
{ "rms", "set the RMS value", OFFSET(rms_value), AV_OPT_TYPE_DOUBLE, {.dbl=0.0}, 0.0, 1.0, FLAGS },
{ "m", "set the RMS value", OFFSET(rms_value), AV_OPT_TYPE_DOUBLE, {.dbl=0.0}, 0.0, 1.0, FLAGS },
{ NULL } { NULL }
}; };
@@ -161,12 +166,16 @@ static void consume_pi(ChannelContext *cc, int nb_samples)
} }
} }
static double next_gain(AVFilterContext *ctx, double pi_max_peak, int bypass, double state) static double next_gain(AVFilterContext *ctx, double pi_max_peak, int bypass, double state,
double pi_rms_sum, int pi_size)
{ {
SpeechNormalizerContext *s = ctx->priv; SpeechNormalizerContext *s = ctx->priv;
const double expansion = FFMIN(s->max_expansion, s->peak_value / pi_max_peak);
const double compression = 1. / s->max_compression; const double compression = 1. / s->max_compression;
const int type = s->invert ? pi_max_peak <= s->threshold_value : pi_max_peak >= s->threshold_value; const int type = s->invert ? pi_max_peak <= s->threshold_value : pi_max_peak >= s->threshold_value;
double expansion = FFMIN(s->max_expansion, s->peak_value / pi_max_peak);
if (s->rms_value > DBL_EPSILON)
expansion = FFMIN(expansion, s->rms_value / sqrt(pi_rms_sum / pi_size));
if (bypass) { if (bypass) {
return 1.; return 1.;
@@ -187,13 +196,15 @@ static void next_pi(AVFilterContext *ctx, ChannelContext *cc, int bypass)
av_assert1(cc->pi[start].size > 0); av_assert1(cc->pi[start].size > 0);
av_assert0(cc->pi[start].type > 0 || s->eof); av_assert0(cc->pi[start].type > 0 || s->eof);
cc->pi_size = cc->pi[start].size; cc->pi_size = cc->pi[start].size;
cc->pi_rms_sum = cc->pi[start].rms_sum;
cc->pi_max_peak = cc->pi[start].max_peak; cc->pi_max_peak = cc->pi[start].max_peak;
av_assert1(cc->pi_start != cc->pi_end || s->eof); av_assert1(cc->pi_start != cc->pi_end || s->eof);
start++; start++;
if (start >= MAX_ITEMS) if (start >= MAX_ITEMS)
start = 0; start = 0;
cc->pi_start = start; cc->pi_start = start;
cc->gain_state = next_gain(ctx, cc->pi_max_peak, bypass, cc->gain_state); cc->gain_state = next_gain(ctx, cc->pi_max_peak, bypass, cc->gain_state,
cc->pi_rms_sum, cc->pi_size);
} }
} }
@@ -209,7 +220,8 @@ static double min_gain(AVFilterContext *ctx, ChannelContext *cc, int max_size)
while (size <= max_size) { while (size <= max_size) {
if (idx == cc->pi_end) if (idx == cc->pi_end)
break; break;
gain_state = next_gain(ctx, cc->pi[idx].max_peak, 0, gain_state); gain_state = next_gain(ctx, cc->pi[idx].max_peak, 0, gain_state,
cc->pi[idx].rms_sum, cc->pi[idx].size);
min_gain = FFMIN(min_gain, gain_state); min_gain = FFMIN(min_gain, gain_state);
size += cc->pi[idx].size; size += cc->pi[idx].size;
idx++; idx++;
@@ -236,11 +248,13 @@ static void analyze_channel_## name (AVFilterContext *ctx, ChannelContext *cc,
\ \
while (n < nb_samples) { \ while (n < nb_samples) { \
ptype new_max_peak; \ ptype new_max_peak; \
ptype new_rms_sum; \
int new_size; \ int new_size; \
\ \
if ((cc->state != (src[n] >= zero)) || \ if ((cc->state != (src[n] >= zero)) || \
(pi[pi_end].size > max_period)) { \ (pi[pi_end].size > max_period)) { \
ptype max_peak = pi[pi_end].max_peak; \ ptype max_peak = pi[pi_end].max_peak; \
ptype rms_sum = pi[pi_end].rms_sum; \
int state = cc->state; \ int state = cc->state; \
\ \
cc->state = src[n] >= zero; \ cc->state = src[n] >= zero; \
@@ -251,10 +265,13 @@ static void analyze_channel_## name (AVFilterContext *ctx, ChannelContext *cc,
pi_end++; \ pi_end++; \
if (pi_end >= MAX_ITEMS) \ if (pi_end >= MAX_ITEMS) \
pi_end = 0; \ pi_end = 0; \
if (cc->state != state) \ if (cc->state != state) { \
pi[pi_end].max_peak = DBL_MIN; \ pi[pi_end].max_peak = DBL_MIN; \
else \ pi[pi_end].rms_sum = 0.0; \
} else { \
pi[pi_end].max_peak = max_peak; \ pi[pi_end].max_peak = max_peak; \
pi[pi_end].rms_sum = rms_sum; \
} \
pi[pi_end].type = 0; \ pi[pi_end].type = 0; \
pi[pi_end].size = 0; \ pi[pi_end].size = 0; \
av_assert1(pi_end != cc->pi_start); \ av_assert1(pi_end != cc->pi_start); \
@@ -262,10 +279,12 @@ static void analyze_channel_## name (AVFilterContext *ctx, ChannelContext *cc,
} \ } \
\ \
new_max_peak = pi[pi_end].max_peak; \ new_max_peak = pi[pi_end].max_peak; \
new_rms_sum = pi[pi_end].rms_sum; \
new_size = pi[pi_end].size; \ new_size = pi[pi_end].size; \
if (cc->state) { \ if (cc->state) { \
while (src[n] >= zero) { \ while (src[n] >= zero) { \
new_max_peak = FFMAX(new_max_peak, src[n]); \ new_max_peak = FFMAX(new_max_peak, src[n]); \
new_rms_sum += src[n] * src[n]; \
new_size++; \ new_size++; \
n++; \ n++; \
if (n >= nb_samples) \ if (n >= nb_samples) \
@@ -274,6 +293,7 @@ static void analyze_channel_## name (AVFilterContext *ctx, ChannelContext *cc,
} else { \ } else { \
while (src[n] < zero) { \ while (src[n] < zero) { \
new_max_peak = FFMAX(new_max_peak, -src[n]); \ new_max_peak = FFMAX(new_max_peak, -src[n]); \
new_rms_sum += src[n] * src[n]; \
new_size++; \ new_size++; \
n++; \ n++; \
if (n >= nb_samples) \ if (n >= nb_samples) \
@@ -282,6 +302,7 @@ static void analyze_channel_## name (AVFilterContext *ctx, ChannelContext *cc,
} \ } \
\ \
pi[pi_end].max_peak = new_max_peak; \ pi[pi_end].max_peak = new_max_peak; \
pi[pi_end].rms_sum = new_rms_sum; \
pi[pi_end].size = new_size; \ pi[pi_end].size = new_size; \
} \ } \
cc->pi_end = pi_end; \ cc->pi_end = pi_end; \