Hello,
Benjamin Larsson a écrit :
> What you have is this issue:
>
>
https://roundup.ffmpeg.org/roundup/ffmpeg/issue271
According to msg2405 (from michaelni):
> A clean fix first requires the audio API to be changed to pass
> AVPackets around instead of buf+size, the nellymoser decoder
> can then insert silence where there are jumps in the dts.
It seems the first part has already been done, thanks to commit [18351].
Benjamin Larsson a écrit :
> Sebastian Wagner wrote:
>> => I understand that now thanks. Although bit curieuse what they did at
>> Adobe.
>> Anyway from that point of view it must be then sufficient to repeat
>> the last
>> null-packets n-times again each time you detect a gap/silence in the
>> Audio
>> channel ... but only if there is a stable calculation between the
>> length of
>> milliseconds of a nellymoser packet and the packet size (44100 Khz is
>> 512Bytes and seem to be approximate 50ms).
>>
>
> The proper way to do it is to use the DTS timestamp (should be found in
> the AVPacket). The nellymoser decoder can then keep track of the DTS and
> prepad the output when a jump is detected. Any other hack will be rejected.
Like what does the attached patch ?
The patch works and is probably good enough for OpenMeetings.
But I guess Sebastian needs a fix clean enough to be committed.
One value is hardcoded (1000). Where can I get it ? I think it's
'(AVFormatContext*)->streams[1]->time_base' (in flv demuxer).
I haven't tested yet a .flv without any gap, to check that the patch doesn't insert any silence.
Why not inserting silences in avcodec_decode_audio3 so that all codecs benefit from it ?
Regards,
Julien
(oops, only members can post - subscribing to ffmpeg-user and resending - sorry for the noise)
diff --git a/libavcodec/nellymoserdec.c b/libavcodec/nellymoserdec.c
index 38f4b42..7fea6c9 100644
--- a/libavcodec/nellymoserdec.c
+++ b/libavcodec/nellymoserdec.c
@@ -49,6 +49,7 @@ typedef struct NellyMoserDecodeContext {
GetBitContext gb;
int add_bias;
float scale_bias;
+ int64_t sample_count;
DSPContext dsp;
FFTContext imdct_ctx;
DECLARE_ALIGNED_16(float,imdct_out)[NELLY_BUF_LEN * 2];
@@ -181,6 +182,14 @@ static int decode_tag(AVCodecContext * avctx,
return buf_size;
}
+ s->sample_count += blocks * NELLY_SAMPLES;
+ if (s->sample_count * 1000 < avpkt->dts * avctx->sample_rate) {
+ int n = blocks * NELLY_SAMPLES * sizeof(int16_t);
+ memset(samples, 0, n);
+ *data_size += n;
+ return 0;
+ }
+
for (i=0 ; i<blocks ; i++) {
nelly_decode_block(s, &buf[i*NELLY_BLOCK_LEN], s->float_buf);
s->dsp.float_to_int16(&samples[i*NELLY_SAMPLES], s->float_buf, NELLY_SAMPLES);
_______________________________________________
ffmpeg-user mailing list
[hidden email]
https://lists.mplayerhq.hu/mailman/listinfo/ffmpeg-user
Locked
