Drop in audio or video, get a transcript and subtitles — all in your browser.
Offline Transcriber runs OpenAI's Whisper speech-recognition model entirely inside your web browser using transformers.js. Drag in a recording and it produces a timestamped transcript plus ready-to-use subtitle files. English, Dutch, German and French are supported, alongside dozens of other languages.
The model downloads once and is cached; after that, transcription works with no server and no upload.
Offline Transcriber converts speech in an audio or video file into text. It decodes the audio in your
browser, resamples it for the model, and runs Whisper to produce a transcript broken into timestamped
segments. You can read the result, copy the plain text, or download subtitles as .srt or
.vtt. Everything happens on your own machine — the file you choose is never uploaded
anywhere. The only network use is a one-time download of the model files from the Hugging Face CDN.
Then choose your model, language and acceleration options and press Transcribe.
Pick a model based on how much accuracy you need versus how long you're willing to wait:
| Model | Download | Best for |
|---|---|---|
| Tiny | ~75 MB | Quick drafts, clear speech, fastest results |
| Base | ~145 MB | Good everyday balance (default) |
| Small | ~480 MB | Best multilingual accuracy, slower |
After transcription, use the export buttons above the results:
<video> and the web.| Control | Action |
|---|---|
| Model | Choose Tiny, Base or Small — accuracy versus speed. |
| Language | Auto-detect, or pick English, Dutch, German or French to force that language. |
| Task | Transcribe keeps the spoken language; Translate → English outputs an English translation. |
| Acceleration | Auto uses WebGPU when available, otherwise CPU. You can force WebGPU or CPU (WASM). |
| Transcribe | Start processing the loaded file. |
| Clear | Remove the current file and results. |
The Tiny / Base / Small models are multilingual and recognise dozens of languages. This tool exposes English, Dutch, German and French directly, plus an Auto-detect option. Forcing the correct language usually gives cleaner results than Auto-detect, especially on short or noisy clips. The Translate → English task will translate speech in any supported language into English text.
Results appear in two tabs. Segments shows each line of speech with its start and end time — this is what becomes your subtitles. Plain text shows the full transcript as one flowing block. A summary line reports how many segments were found, the total duration, and which model was used.
If your browser supports WebGPU (recent Chrome, Edge, and Chromium-based browsers), transcription runs on your GPU and is several times faster. Otherwise it falls back to CPU via WebAssembly, which still works but is slower — pick the Tiny model for long files on CPU. The control panel shows whether WebGPU was detected. Processing happens in a background worker, so the page stays responsive while it runs.
Offline Transcriber processes everything locally in your browser. The audio or video you load is decoded
and transcribed on your device and is never transmitted to any server. The only network
request the tool makes is a one-time download of the Whisper model files from a public CDN; after that the
tool works without re-downloading. Your settings (model, language, task, acceleration) are stored in
localStorage on your device and never leave it.
Common audio: .mp3 · .wav · .m4a · .aac · .ogg · .flac
Video (audio track is extracted): .mp4 · .webm · .mov
Technologies used:
Built with: