[FFmpeg-user] Stuck on adelay/amix sync problem..

[Paul B Mahol]

On 12/19/17, Mark Corner <mcorner at gmail.com> wrote:
> I am having trouble with amix producing different results with different
> input orders and only one input to amix staying in sync with the video.
>
> I am starting with three video and three audio files, vp8 and opus
> encoded.  There is a video and audio that go together and are about  32
> minutes long.  Then there are two pairs of audio and video, each about 16
> minutes long.
>
> First I combine the three videos using an overlay:
> ffmpeg -y -itsoffset 0 -i
> /tmp/RM359a097bf459c327e34954525862d2b7/RTd9075ac090a32a21e71269e6b08f4053-reproces
> sed.mkv -itsoffset 2.364 -i
> /tmp/RM359a097bf459c327e34954525862d2b7/RT21a59c19d91a7644b8c76d3fd12c3b32-reproc
> essed.mkv -itsoffset 983.503 -i
> /tmp/RM359a097bf459c327e34954525862d2b7/RT4d34aa2bae28a73f8079a6e1b65b3aa4-re
> processed.mkv  -filter_complex
> "[0]pad=iw*2:ih[t0];[t0][1]overlay=W/2:0[t1];[t1][2]overlay=0:0[t2]" -map
> [t2]
>  -an -c:v libvpx -crf 23 -b:v 1M -cpu-used 3 -threads 8
> /tmp/RM359a097bf459c327e34954525862d2b7/output_video.
> webm
>
> That works fine.
>
> Then I combine the audio (this is where the problem is):
> ffmpeg -y -i
> /tmp/RM359a097bf459c327e34954525862d2b7/RT2568dcb62e48c44c4ac61094c4377553
> -i /tmp/RM359a097bf45
> 9c327e34954525862d2b7/RTb5d9c3c0a59e98061159ddb310587b71 -i
> /tmp/RM359a097bf459c327e34954525862d2b7/RT955f8ed
> 764b5fda55a4dcd23cfed673d  -filter_complex
> "[0]adelay=25|25[t0];[1]adelay=2363|2363[t1];[2]adelay=983525|9835
> 25[t2];[t2][t1][t0]amix=inputs=3" -strict -2 -c:a libopus
> /tmp/RM359a097bf459c327e34954525862d2b7/output_audi
> o.mka
>
> And then merge the audio and video together:
> ffmpeg -y -i /tmp/RM359a097bf459c327e34954525862d2b7/output_audio.mka -i
> /tmp/RM359a097bf459c327e34954525862d2b7/output_video.webm -c:v copy -c:a
> copy /tmp/RM359a097bf459c327e34954525862d2b7/full.webm
>
> The problem is that the audio and video are only in sync for what ever is
> first in the amix filter....  So in the above example, the t2 audio input
> will be in sync, but not the t1 or t0.  If I change the order to something
> like [t1][t2][t0], then the t1 audio is in sync, but none of the others.
> In fact the length of the output_audio.mka will differ based on the order
> of inputs to amix....  As far as I understand, the order to amix shouldn't
> matter at all.
>
> Thanks for any advice you can offer.

Perhaps you need to use asetpts filter after each adelay to reset pts
to sane value?