What is dual-anchor-audio-sync?

Dual Anchor Dub Sync Pipeline is a set of scripts, not a finished app, that solves a very specific video problem: lining up a foreign language dub of a TV show with a better quality original language master that runs at a different frame rate. It was built around Star Trek Deep Space Nine, matching a German DVD rip running at PAL speed to a higher quality English laserdisc upscale, but the technique itself is not tied to that show. The clever part of the approach is that both the German source and the English source usually contain an English audio track. Rather than trying to match German audio against English audio directly, which is unreliable, the pipeline matches English against English using sample accurate cross correlation, then applies that same timing correction to the German track. This timing correction is not just a simple constant speed adjustment for the frame rate difference, it also accounts for scene cuts that differ between versions, such as censorship changes or different intro and outro lengths, where the offset actually jumps rather than drifting smoothly. The pipeline detects these jumps and places the dubbed audio onto the timeline piece by piece, filling any gaps with silence. On a real ninety minute episode, this method found eight such edit points and brought the leftover timing drift down to sixteen milliseconds, compared to over five and a half seconds of drift beforehand. The toolkit is a shell script that calls out to smaller Python scripts and a few well known audio and video tools, and it produces a single combined video file with both audio tracks and subtitles included. German subtitles have one known limitation: because they only exist as bitmap images rather than text, only a simple linear timing correction can be applied to them rather than the more precise piecewise correction used for audio. This project would suit hobbyists working on personal media preservation or restoration projects who need to fix audio sync issues between mismatched language versions of the same recording, and it requires no copyrighted media of its own, only scripts and documentation. It is released under the MIT license.