Transcribe
Transcribe writes a subtitle file for every video you add, from the speech in it. It runs entirely on your computer — your videos are never uploaded — and saves each result next to its video.
Adding videos
Click Add videos, or drop video files or a whole folder anywhere on the window. A folder is scanned and every video inside is added. Each file is read first, then shows its length and size and is ticked automatically.
The search box in the header filters the list by file name. Check all and Uncheck all tick every file at once, and Remove all empties the list. Removing a file that is queued or running asks first, because it stops that job.
Transcribe and Extract each keep their own list, so adding a season here to pull subtitles out of it does not queue it for transcription. A file with no subtitle tracks on the Extract screen has a Transcribe instead button that hands it across.
Choosing the language and the speech engine
Two dropdowns in the toolbar decide what the AI listens for:
- Language spoken in the video — the language of the dialogue. Whisper needs to be told. For an engine that detects the language itself, this shows Auto-detect.
- Speech model — Whisper models you have downloaded (Tiny to Large; larger is more accurate and slower), plus Qwen3-ASR and Parakeet once installed. An engine you have not installed is still listed, with its download size; choosing it opens an install dialog and selects it when the download finishes.
See Speech Engines for what each engine is good at. The language and model you pick here start as the defaults set in Settings → General.
Vocabulary hint
The Vocabulary field takes names and terms spoken in the audio — for example Goku, Vegeta, Saiyan — so they are spelled correctly. Use names that are spoken, not the title of the show or episode, which Whisper may repeat. It is limited to 900 characters.
Suggest fills it in with AI: from the checked files’ names, or from a title you type. It looks up the characters and places of the show or film and replaces what is in the field. Suggesting needs a DeepSeek key (Settings → AI Providers).
The hint works for Whisper and Qwen3-ASR; Parakeet does not use it. With a hint, Whisper reads it as context, which helps spelling but can pull its timing onto a fixed one-to-two-second grid. If you see that, turn on Reduce hallucinations in Settings → General.
Running it
Tick the files you want and click Transcribe (it shows how many are ticked). A file shows Transcribing… with progress, then Done or Failed. If the machine is busy a file shows Waiting until it has its share of the processor or graphics card. Stop cancels a running file and Unqueue takes a waiting one out.
Each subtitle is saved beside its video as Name.language.srt — for example Film.en.srt. When a file finishes, Edit opens it in Edit Subtitles and Show in folder reveals it. A failed file keeps its error message and a Try again button, and results stay on screen until you run the file again or remove it.
You can leave the screen while files run — jobs continue in the background, and the app asks before quitting during one.
Several videos at once
Videos transcribed at once in Settings → General (1 to 4, default 1) sets how many files run together. The app divides your processor threads between the running files, and a lone file always gets all of them however high the setting is. With a graphics-card build of Whisper the files run one at a time and the setting is disabled, because two copies of the model in video memory are not worth the small gain.
Other ways to start
Right-clicking a video in Explorer or Finder offers Transcribe and Extract subtitles beside Open with. The app is offered for videos but never made the default player. To speak into a microphone instead of a file, see Live transcription.