The itch
The best educational content on YouTube — math, programming, science — is overwhelmingly in English. For Sinhala speakers, that language barrier turns a 20-minute lesson into a struggle. YouTube’s auto-translated captions come out as word-salad, and paid dubbing services don’t cover Sinhala at all.
Vidura’s intention is simple: make English educational video genuinely watchable for Sinhala speakers, with translations that read like something a person would actually say — and timestamps accurate to the frame.
How it works
Vidura fetches a video’s own caption track (via yt-dlp), then translates the whole transcript in a single structured-output LLM call with full video context — so terminology stays consistent and sentences flow naturally across subtitle lines. The synced result overlays the embedded player with VLC-style subtitle controls.
Every video wears its provenance openly: where the timestamps came from, a computed sync-quality score, and which model translated it.
Beyond subtitles
- Transcript-grounded chat — ask questions about one video or your whole library; answers cite the video and timestamp
- Timestamped notes pinned to exact moments
- Installable PWA with offline shell, live processing progress over SSE, and web push when a video finishes
- Self-hosted — your library, notes, and chat history live on your server
- Any target language — Sinhala is the default, not the limit
Status
Actively maintained. Source, setup guide, and screenshots on GitHub.