[FFmpeg-cvslog] avfilter/af_loudnorm: fix filtering of last 2.9 seconds
Paul B Mahol
git at videolan.org
Wed Feb 23 18:01:03 EET 2022
ffmpeg | branch: master | Paul B Mahol <onemda at gmail.com> | Wed Feb 23 10:20:58 2022 +0100| [57f0cdbe17dfe5b304898aa6d05ab9df4bdb284d] | committer: Paul B Mahol
avfilter/af_loudnorm: fix filtering of last 2.9 seconds
> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=57f0cdbe17dfe5b304898aa6d05ab9df4bdb284d
---
libavfilter/af_loudnorm.c | 131 +++++++++++++---------------------------------
1 file changed, 36 insertions(+), 95 deletions(-)
diff --git a/libavfilter/af_loudnorm.c b/libavfilter/af_loudnorm.c
index 9bb0c65bb7..493306c707 100644
--- a/libavfilter/af_loudnorm.c
+++ b/libavfilter/af_loudnorm.c
@@ -408,37 +408,45 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
AVFilterContext *ctx = inlink->dst;
LoudNormContext *s = ctx->priv;
AVFilterLink *outlink = ctx->outputs[0];
+ const int final_samples = FFMIN(19200, inlink->sample_count_out - outlink->sample_count_in);
AVFrame *out;
- const double *src;
+ const double *src = NULL;
double *dst;
double *buf;
double *limiter_buf;
- int i, n, c, subframe_length, src_index;
+ int n, c, subframe_length;
double gain, gain_next, env_global, env_shortterm,
global, shortterm, lra, relative_threshold;
- if (av_frame_is_writable(in)) {
- out = in;
- } else {
- out = ff_get_audio_buffer(outlink, in->nb_samples);
+ if (in) {
+ if (s->frame_type == FIRST_FRAME && in->nb_samples < frame_size(inlink->sample_rate, 3000))
+ s->frame_type = LINEAR_MODE;
+
+ out = ff_get_audio_buffer(outlink, s->frame_type == LINEAR_MODE ? in->nb_samples : 19200);
if (!out) {
av_frame_free(&in);
return AVERROR(ENOMEM);
}
av_frame_copy_props(out, in);
+ } else {
+ out = ff_get_audio_buffer(outlink, 19200);
+ if (!out)
+ return AVERROR(ENOMEM);
}
out->pts = s->pts[0];
memmove(s->pts, &s->pts[1], (FF_ARRAY_ELEMS(s->pts) - 1) * sizeof(s->pts[0]));
- src = (const double *)in->data[0];
dst = (double *)out->data[0];
buf = s->buf;
limiter_buf = s->limiter_buf;
- ff_ebur128_add_frames_double(s->r128_in, src, in->nb_samples);
+ if (in) {
+ src = (const double *)in->data[0];
+ ff_ebur128_add_frames_double(s->r128_in, src, in->nb_samples);
+ }
- if (s->frame_type == FIRST_FRAME && in->nb_samples < frame_size(inlink->sample_rate, 3000)) {
+ if (s->frame_type == FIRST_FRAME && in && in->nb_samples < frame_size(inlink->sample_rate, 3000)) {
double offset, offset_tp, true_peak;
ff_ebur128_loudness_global(s->r128_in, &global);
@@ -452,7 +460,6 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
offset = pow(10., (s->target_i - global) / 20.);
offset_tp = true_peak * offset;
s->offset = offset_tp < s->target_tp ? offset : s->target_tp - true_peak;
- s->frame_type = LINEAR_MODE;
}
switch (s->frame_type) {
@@ -502,16 +509,19 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
s->frame_type = INNER_FRAME;
break;
+ case FINAL_FRAME:
case INNER_FRAME:
gain = gaussian_filter(s, s->index + 10 < 30 ? s->index + 10 : s->index + 10 - 30);
gain_next = gaussian_filter(s, s->index + 11 < 30 ? s->index + 11 : s->index + 11 - 30);
- for (n = 0; n < in->nb_samples; n++) {
+ for (n = 0; n < out->nb_samples; n++) {
for (c = 0; c < inlink->channels; c++) {
- buf[s->prev_buf_index + c] = src[c];
- limiter_buf[s->limiter_buf_index + c] = buf[s->buf_index + c] * (gain + (((double) n / in->nb_samples) * (gain_next - gain))) * s->offset;
+ if (src)
+ buf[s->prev_buf_index + c] = src[c];
+ limiter_buf[s->limiter_buf_index + c] = buf[s->buf_index + c] * (gain + (((double) n / out->nb_samples) * (gain_next - gain))) * s->offset;
}
- src += inlink->channels;
+ if (src)
+ src += inlink->channels;
s->limiter_buf_index += inlink->channels;
if (s->limiter_buf_index >= s->limiter_buf_size)
@@ -526,11 +536,11 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
s->buf_index -= s->buf_size;
}
- subframe_length = (frame_size(inlink->sample_rate, 100) - in->nb_samples) * inlink->channels;
+ subframe_length = (frame_size(inlink->sample_rate, 100) - out->nb_samples) * inlink->channels;
s->limiter_buf_index = s->limiter_buf_index + subframe_length < s->limiter_buf_size ? s->limiter_buf_index + subframe_length : s->limiter_buf_index + subframe_length - s->limiter_buf_size;
- true_peak_limiter(s, dst, in->nb_samples, inlink->channels);
- ff_ebur128_add_frames_double(s->r128_out, dst, in->nb_samples);
+ true_peak_limiter(s, dst, out->nb_samples, inlink->channels);
+ ff_ebur128_add_frames_double(s->r128_out, dst, out->nb_samples);
ff_ebur128_loudness_range(s->r128_in, &lra);
ff_ebur128_loudness_global(s->r128_in, &global);
@@ -560,51 +570,9 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
s->index++;
if (s->index >= 30)
s->index -= 30;
- s->prev_nb_samples = in->nb_samples;
- break;
-
- case FINAL_FRAME:
- gain = gaussian_filter(s, s->index + 10 < 30 ? s->index + 10 : s->index + 10 - 30);
- s->limiter_buf_index = 0;
- src_index = 0;
-
- for (n = 0; n < s->limiter_buf_size / inlink->channels; n++) {
- for (c = 0; c < inlink->channels; c++) {
- s->limiter_buf[s->limiter_buf_index + c] = src[src_index + c] * gain * s->offset;
- }
- src_index += inlink->channels;
-
- s->limiter_buf_index += inlink->channels;
- if (s->limiter_buf_index >= s->limiter_buf_size)
- s->limiter_buf_index -= s->limiter_buf_size;
- }
-
- subframe_length = frame_size(inlink->sample_rate, 100);
- for (i = 0; i < in->nb_samples / subframe_length; i++) {
- true_peak_limiter(s, dst, subframe_length, inlink->channels);
-
- for (n = 0; n < subframe_length; n++) {
- for (c = 0; c < inlink->channels; c++) {
- if (src_index < (in->nb_samples * inlink->channels)) {
- limiter_buf[s->limiter_buf_index + c] = src[src_index + c] * gain * s->offset;
- } else {
- limiter_buf[s->limiter_buf_index + c] = 0.;
- }
- }
-
- if (src_index < (in->nb_samples * inlink->channels))
- src_index += inlink->channels;
-
- s->limiter_buf_index += inlink->channels;
- if (s->limiter_buf_index >= s->limiter_buf_size)
- s->limiter_buf_index -= s->limiter_buf_size;
- }
-
- dst += (subframe_length * inlink->channels);
- }
-
- dst = (double *)out->data[0];
- ff_ebur128_add_frames_double(s->r128_out, dst, in->nb_samples);
+ if (s->frame_type == FINAL_FRAME)
+ out->nb_samples = final_samples;
+ s->prev_nb_samples = out->nb_samples;
break;
case LINEAR_MODE:
@@ -617,11 +585,12 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
}
dst = (double *)out->data[0];
+ out->nb_samples = in->nb_samples;
ff_ebur128_add_frames_double(s->r128_out, dst, in->nb_samples);
break;
}
- if (in != out)
+ if (in)
av_frame_free(&in);
return ff_filter_frame(outlink, out);
}
@@ -634,38 +603,9 @@ static int flush_frame(AVFilterLink *outlink)
int ret = 0;
if (s->frame_type == INNER_FRAME) {
- double *src;
- double *buf;
- int nb_samples, n, c, offset;
- AVFrame *frame;
-
- nb_samples = (s->buf_size / inlink->channels) - s->prev_nb_samples;
- nb_samples -= (frame_size(inlink->sample_rate, 100) - s->prev_nb_samples);
-
- frame = ff_get_audio_buffer(outlink, nb_samples);
- if (!frame)
- return AVERROR(ENOMEM);
- frame->nb_samples = nb_samples;
-
- buf = s->buf;
- src = (double *)frame->data[0];
-
- offset = ((s->limiter_buf_size / inlink->channels) - s->prev_nb_samples) * inlink->channels;
- offset -= (frame_size(inlink->sample_rate, 100) - s->prev_nb_samples) * inlink->channels;
- s->buf_index = s->buf_index - offset < 0 ? s->buf_index - offset + s->buf_size : s->buf_index - offset;
-
- for (n = 0; n < nb_samples; n++) {
- for (c = 0; c < inlink->channels; c++) {
- src[c] = buf[s->buf_index + c];
- }
- src += inlink->channels;
- s->buf_index += inlink->channels;
- if (s->buf_index >= s->buf_size)
- s->buf_index -= s->buf_size;
- }
-
s->frame_type = FINAL_FRAME;
- ret = filter_frame(inlink, frame);
+ while (inlink->sample_count_out > outlink->sample_count_in)
+ ret = filter_frame(inlink, NULL);
}
return ret;
}
@@ -712,8 +652,9 @@ static int activate(AVFilterContext *ctx)
return ret;
if (ff_inlink_acknowledge_status(inlink, &status, &pts)) {
+ ret = flush_frame(outlink);
ff_outlink_set_status(outlink, status, pts);
- return flush_frame(outlink);
+ return ret;
}
FF_FILTER_FORWARD_WANTED(outlink, inlink);
More information about the ffmpeg-cvslog
mailing list