Gronis
commented
2 years ago Are you thinking about syncing using a reference audio stream with another audio stream (for example a language dub), or are you looking at syncing audio using a video reference (like reading lip motion or something)? The first sounds absolutely feasible. The second sounds more like a research subject and not that straight forward.
To my understanding, most of the machine learning part un subsync is library based and not developed as part of the application. You can read more at @sc0ty's blog: https://sc0ty.pl/2019/04/subsync-architecture-overview/
Hello
As per the title, this isnt an issue I was just curious about the machine learning aspects of Subsync and whether or not anything like this could be applied to audio/video sync?
Obviously its speech recognition with txt subtitles vs audio/video sync which would require speech recognition coupled with facial/lip movement matching. So a lot more difficult. I suspect anyone who could do it would be charging money for it.
But after recently grappling with some out of sync files and being able to fix a bunch of subs automagically using Subsync but having to manually fiddle with the audio I was curious if a super dooper magic audio/video sync tool was even technically possible.
Anyway thanks for the great sub sync tool, sure beats the bad old days of virtualdub and its ilk.