Audio transcriber
Turn a recording or video into text, SRT, or WebVTT subtitles. Runs entirely in your browser — your audio is never uploaded, there is no per-minute charge, and there is no limit on how much you transcribe.
About this transcriber
This turns spoken audio into text using Whisper, a speech recognition model from OpenAI released under a permissive licence. It handles recordings and video files alike, and exports either a readable transcript or subtitle files you can attach to a video.
Your recording is never uploaded. The model is downloaded to your browser once, then everything runs on your own machine. That is the real difference from cloud transcription services, which charge per minute and take a copy of your audio in the process. Interviews, therapy notes, legal recordings, and anything else sensitive stay where they are.
Because the work happens on your hardware rather than ours, there is no per-minute cost to us and therefore no usage cap, no watermark, and no account. Speed depends on your machine: a browser with GPU support is many times faster than one falling back to the CPU.
Common questions
- Is my audio uploaded anywhere?
- No. The speech model is downloaded to your browser on first use, and after that the transcription runs entirely on your device. You can disconnect from the network after the model loads and it still works.
- How do I make subtitles for a video?
- Drop the video file in directly — the audio track is extracted for you. Choose SRT for YouTube and most players, or WebVTT for web video, then download. Both include the timings.
- How long can the recording be?
- Up to about 30 minutes. Whisper processes audio in 30-second windows and each one is a full model pass held in browser memory, so this is a genuine ceiling rather than a throttle. Split longer recordings, or use a desktop app.
- Why is the first run slow?
- The model file is around 60MB and downloads once, then your browser caches it. Subsequent transcriptions skip that entirely and start immediately.
- How accurate is it?
- Good on clear speech in English, noticeably worse with heavy background noise, crosstalk, or strong accents. It does not identify who is speaking. Treat the output as a strong first draft to correct rather than a finished document.
- Does it work on other languages?
- Not currently — this uses the English-only model, which is smaller and more accurate on English than the multilingual version of the same size.