af_resample: do not touch the timestamps if we are not resampling

This filter currently assumes that the input audio is continuous and
does some timestamps manipulation based on this assumption.

This is unnecessary if we are only converting the channel layout or the
sample format, without resampling. In such a case, just leave the
timestamps as they are.
This commit is contained in:
Anton Khirnov 2015-07-16 19:26:36 +02:00
parent 6d592fbd0d
commit 6b15874fc2

View File

@ -41,6 +41,7 @@ typedef struct ResampleContext {
AVAudioResampleContext *avr;
AVDictionary *options;
int resampling;
int64_t next_pts;
int64_t next_in_pts;
@ -118,6 +119,8 @@ static int config_output(AVFilterLink *outlink)
char buf1[64], buf2[64];
int ret;
int64_t resampling_forced;
if (s->avr) {
avresample_close(s->avr);
avresample_free(&s->avr);
@ -156,9 +159,15 @@ static int config_output(AVFilterLink *outlink)
if ((ret = avresample_open(s->avr)) < 0)
return ret;
outlink->time_base = (AVRational){ 1, outlink->sample_rate };
s->next_pts = AV_NOPTS_VALUE;
s->next_in_pts = AV_NOPTS_VALUE;
av_opt_get_int(s->avr, "force_resampling", 0, &resampling_forced);
s->resampling = resampling_forced || (inlink->sample_rate != outlink->sample_rate);
if (s->resampling) {
outlink->time_base = (AVRational){ 1, outlink->sample_rate };
s->next_pts = AV_NOPTS_VALUE;
s->next_in_pts = AV_NOPTS_VALUE;
} else
outlink->time_base = inlink->time_base;
av_get_channel_layout_string(buf1, sizeof(buf1),
-1, inlink ->channel_layout);
@ -241,7 +250,7 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
av_assert0(!avresample_available(s->avr));
if (s->next_pts == AV_NOPTS_VALUE) {
if (s->resampling && s->next_pts == AV_NOPTS_VALUE) {
if (in->pts == AV_NOPTS_VALUE) {
av_log(ctx, AV_LOG_WARNING, "First timestamp is missing, "
"assuming 0.\n");
@ -260,22 +269,25 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
goto fail;
}
out->sample_rate = outlink->sample_rate;
/* Only convert in->pts if there is a discontinuous jump.
This ensures that out->pts tracks the number of samples actually
output by the resampler in the absence of such a jump.
Otherwise, the rounding in av_rescale_q() and av_rescale()
causes off-by-1 errors. */
if (in->pts != AV_NOPTS_VALUE && in->pts != s->next_in_pts) {
out->pts = av_rescale_q(in->pts, inlink->time_base,
outlink->time_base) -
av_rescale(delay, outlink->sample_rate,
inlink->sample_rate);
} else
out->pts = s->next_pts;
if (s->resampling) {
out->sample_rate = outlink->sample_rate;
/* Only convert in->pts if there is a discontinuous jump.
This ensures that out->pts tracks the number of samples actually
output by the resampler in the absence of such a jump.
Otherwise, the rounding in av_rescale_q() and av_rescale()
causes off-by-1 errors. */
if (in->pts != AV_NOPTS_VALUE && in->pts != s->next_in_pts) {
out->pts = av_rescale_q(in->pts, inlink->time_base,
outlink->time_base) -
av_rescale(delay, outlink->sample_rate,
inlink->sample_rate);
} else
out->pts = s->next_pts;
s->next_pts = out->pts + out->nb_samples;
s->next_in_pts = in->pts + in->nb_samples;
s->next_pts = out->pts + out->nb_samples;
s->next_in_pts = in->pts + in->nb_samples;
} else
out->pts = in->pts;
ret = ff_filter_frame(outlink, out);
s->got_output = 1;