Thanks Ryan! After working on this for far too long I discovered a solution just before going crazy:
Run ashowinfo and parse out the first audio packet Use aevalsrc 0 to generate silence exactly as long as time of first audio packet. Concat that silence on the front of the audio stream. This fixes the offset issue when doing any amix down the road, however it can cause issues with amerge (mostly warnings) that I still haven't gotten to the bottom of. Best, Ned On Wednesday, December 9, 2015, Ryan Williams <[email protected]> wrote: > I'm not sure if this will be of any help but look into the asetpts, adelay > and amix filters. > You may also find that the afifo filter may help but be careful as it > consumes large amounts of memory. > > Failing that; perhaps you could generate a stream of silence with > aevalsrc='0|0' then merge each stream with silence prior to merging them > together. > This would at least give you two audio streams with samples commencing > from 00:00:00.00 > > -----Original Message----- > From: ffmpeg-user [mailto:[email protected] <javascript:;>] > On Behalf Of Ned Rockson > Sent: Thursday, 10 December 2015 07:06 > To: [email protected] <javascript:;> > Subject: [FFmpeg-user] Compensating for missing audio frames at beginning > of video > > Hi, > > I have a problem with audio syncing when using the amerge filter on two > video streams where the audio may be missing for an unknown period of time > on the second stream. To use slightly more concrete terms, I have a simple > example: > > * Two videos of the same length, same encoding, same audio sampling rates. > * Using ashowinfo and inspecting the first packet the difference becomes > apparent: > > ===First video=== > ffmpeg -i a_re.webm -af 'ashowinfo' -f null /dev/null ... > [Parsed_ashowinfo_0 @ 0x7fe7f8c19de0] n:0 pts:79584 pts_time:1.658 > pos:7827 fmt:fltp channels:1 chlayout:mono rate:48000 nb_samples:960 > checksum:DE7F2A43 plane_checksums: [ DE7F2A43 ] > [Parsed_ashowinfo_0 @ 0x7fe7f8c19de0] n:1 pts:80544 pts_time:1.678 > pos:7902 fmt:fltp channels:1 chlayout:mono rate:48000 nb_samples:960 > checksum:15BE5F14 plane_checksums: [ 15BE5F14 ] .... > > ===Second video=== > ffmpeg -i b_re.webm -af 'ashowinfo' -f null /dev/null ... > [Parsed_ashowinfo_0 @ 0x7f8711700000] n:0 pts:192 pts_time:0.004 pos:8537 > fmt:fltp channels:1 chlayout:mono rate:48000 nb_samples:960 > checksum:2E9F59AA plane_checksums: [ 2E9F59AA ] > [Parsed_ashowinfo_0 @ 0x7f8711700000] n:1 pts:1152 pts_time:0.024 > pos:19246 fmt:fltp channels:1 chlayout:mono rate:48000 nb_samples:960 > checksum:68193A05 plane_checksums: [ 68193A05 ] .... > > The amerge takes each frame and merges them, one by one based on the > STARTPTS of input 0. I'm curious if there's any way to make this have an > offset for the audio in the videos to the STARTPTS for that respective > audio? > > Best, > Ned > > -- > Ned Rockson > _______________________________________________ > ffmpeg-user mailing list > [email protected] <javascript:;> > http://ffmpeg.org/mailman/listinfo/ffmpeg-user > > _______________________________________________ > ffmpeg-user mailing list > [email protected] <javascript:;> > http://ffmpeg.org/mailman/listinfo/ffmpeg-user > -- Ned Rockson m: +1 415 580 0074 | @nedrockson _______________________________________________ ffmpeg-user mailing list [email protected] http://ffmpeg.org/mailman/listinfo/ffmpeg-user
