Notes on timed voiceover from the team behind Recastr: why dubs drift, what subtitle files really contain, and how to re-voice a recording without touching the edit.
Two kinds of separator, given narration over songs with lyrics. The music model put the singer in the speech stem every time. The cinematic one mostly did not.
· 7 min read · separation, background-audio
Eight speech-to-text models, ten runs each, the same eight clips. The most accurate, the cheapest and the one that can build a timed SRT are not the same model.
· 9 min read · transcription, benchmarks