Show HN: Videolangua – end-to-end video translate and subtitle/ dub

What it is A practical pipeline to go from video → transcripts → bilingual subtitles (A/B at once) or voice dub (choose female or male). It targets lectures, tutorials, tech talks etc

Current constraints (honest version)

No speaker diarization yet → one TTS voice for the entire video

Voice choice: female / male

Bilingual subtitles (e.g., EN+JP, EN+ZH, EN+KR) with length-aware line breaking

You can add customized terms and translations in front of the listening model and the translation model to get the best accuracy

Long videos: we re-anchor timestamps periodically to fight drift

Pipeline (condensed)

ASR with word-level timestamps

Segment cleanup + merge tiny fragments

MT twice → produce A/B subtitle tracks; constrain length to reduce overflow; have multiple language model to supervise/double check the translation quality

TTS → single voice (female/male) for the full track

Mixback → keep ambience, duck original, SRT (mono or dual-lang) + dubbed MP4

Why post this now It’s not “magic studio” quality, but it’s dependable for many real-world cases: course videos, onboarding, webinars. We found being explicit about limits (no diarization) actually speeds teams up.

What we’d love feedback on

The current User experience

Acceptable subtitle overflow rate on 30–60 min content

TTS pacing rules that feel most natural for multilingual reading speed

(If there’s interest, we’ll extract a minimal CLI with the exact steps and corner cases called out.)