Live demo
Subtitle generator
Drop in a video or an audio file and get a cleaned SRT back. Transcription runs on Whisper large-v3 on a GPU in my flat in London, which is also the honest reason for the limits below.
Subtitle generator
Video or audio in, a cleaned SRT out.
How it works
The file is uploaded to a FastAPI service, queued, and transcribed with Whisper. The raw output is then cleaned up: cues are merged and split so they read at a sensible pace, and a terms file corrects the proper nouns Whisper reliably gets wrong.
Nothing is kept. Uploads and generated subtitles are deleted once the job expires, and the file never leaves the machine it was transcribed on.
The queue, the per-visitor limits and the daily ceiling exist because this is one GPU on a domestic connection rather than a cluster. If it is busy, you will be told your position rather than left watching a spinner.