audio synced in avidemux, drifts at end of rendered movie

[mrmcsober]

I bought a movie with a great blu ray video transfer but they badly messed up the audio transfer, it's as though they tried to do a surround mix and chose the wrong tracks.

However there is a dvd version of the same movie (which I bought at the same time because I knew in advance the blu ray had this problem), with a great Dolby 2.0 stereo track.

So the theory went, I will combine the best of both worlds - 1920x1080 video from the blu ray disc, and the Dolby 2.0 stereo from the dvd.

The twist in the tale was that the dvd is PAL, speeded up to 25fps with corresponding slight chipmunkization of the original cinematic audio. However the audio is easy to fix, and I was able to recode the (in both cases 48khz) audio with a 24000/1001/25 correction ratio.

In addition to this the blu ray has exactly 2.5 seconds (120000 audio samples) of blackout at the start. The blu ray video is 152160 frames, the dvd is 152100 frames. In theory the video matches up perfectly after 60 extra frames before the dvd video.

So this command I thought would mux the two together correctly:

ffmpeg -i title_t00.mkv -itsoffset 2.5 -i dvd_audio_new.wav -map 0:v -map 1:a -c:v copy -c:a aac -b:a 192k -f mkv movie_in_sync.mkv

where dvd_audio_new.wav was produced by SoX with the adjusted speed (slight slowdown).

I know this worked correctly because I can load title_t00.mkv and dvd_audio_new.wav into avidemux, add a 2500ms offset to the audio, and they match up perfectly throughout the picture.

But the ffmpeg output file mkv movie_in_sync.mkv is wrong - it starts out in sync, but after about an hour in, the lip sync has slipped by a few frames which of course ruins the dialogue.

So I can only guess there is some kind of marker problem related to keyframes or something that is lost by muxing the two separate files, that somehow is not a problem for avidemux, which perhaps is more tolerant to random seeking than most players.

Can anyone shed light on this or suggest a fix?

[davexnet]

General technique would be to split the audio where it goes out and offset the separated part until it syncs,
and ensure it stays good until the end. If it's good you can rejoin the parts