LiveAI2026
Tafrigh
Transcribe Arabic audio and generate SRT and VTT subtitles.
A transcription tool that turns Arabic audio into text and timed subtitle files using Whisper models and wit.ai. Widely used for making Arabic lectures and books searchable.
- Role
- Contributor
- Year
- 2026
- Status
- Live
- Built with
- Python
- Whisper
- wit.ai
- SRT / VTT
- Where to find it
Arabic speech is under-served by transcription tooling, and the material that most needs it, lectures, lessons, recorded books, is exactly the material nobody has budget to transcribe by hand.
Tafrigh produces both plain text and timed subtitle files, which is what makes the output useful for search rather than just for reading. It is the same capability that Kashaf Abi Ja'far depends on.
Challenges
Timed output is what makes a transcript searchable
The snag
A wall of transcribed text tells you a phrase was said but not where, which leaves the listener scrubbing through the recording anyway.
What I did
Output includes SRT and VTT with segment timings, so downstream tools can index the text and still link back to the exact moment in the audio.
Results
- Plain text plus timed SRT/VTT output for Arabic audio.