Transcribe audio to text
Drop a recording, a voice memo or a video and get its text in your language. The speech recognition runs on your device.
Sealed on this device
Runs in your browser. Nothing is uploaded.
Sealed
Your recording
How it works
-
Add your recording
Drop an audio or video file, or choose it. It is read on your device and never uploaded.
-
Pick the language
Leave “Detect automatically” or choose the spoken language, then start. The first time, the speech model is downloaded once.
-
Copy or download
The text appears as it’s written. Correct it if needed, then copy it or save it as a .txt file.
Frequently asked questions
Is my recording uploaded to a server?
No. The file is decoded and transcribed inside your browser, on your device. Only the speech model is downloaded, from this site, and the browser keeps it for next time. Your audio and its text are never sent anywhere.
How accurate is the transcription?
It is good on clear speech, but not perfect. Background noise, music, strong accents, people talking over each other, names and technical words cause mistakes. Read the text before using it and correct it in place. The tool does not tell speakers apart.
Which languages can it transcribe?
More than 50, including English, Spanish, French, German, Portuguese, Italian, Dutch, Polish, Russian, Ukrainian, Turkish, Arabic, Hindi, Chinese, Japanese, Korean, Indonesian, Vietnamese and Thai. Accuracy is best for widely spoken languages. Automatic detection listens to the first 30 seconds.
How long does it take?
It depends on your device and the length of the recording. The speech recognition runs on your processor, so a recent computer is much faster than an older phone. The first run also includes the one-time model download, after you agree to it. You can cancel at any moment and keep what was already transcribed.
Which files are supported?
Most audio and video files: MP3, WAV, M4A, AAC, OGG, Opus, FLAC, MP4, WebM and MOV. When the browser can’t read a file itself, the tool tries a built-in converter. Files up to 2 GB can be opened; very long recordings need a lot of memory.
Which open-source components does it use?
The speech recognition is Whisper (base model), released by OpenAI under the MIT license, in its ONNX version by the ONNX Community. It runs with Transformers.js by Hugging Face (Apache-2.0 license) and ONNX Runtime Web by Microsoft (MIT license). Files your browser can’t decode itself are converted with FFmpeg compiled to WebAssembly (ffmpeg.wasm), whose core is licensed under the GPL.