Which language is this?
Speak, or upload a recording, and an AI model works out which of 99 languages it is — right in your browser.
Everything runs on your device. Your voice is never uploaded.
Press record and speak for 5–15 secondsAny language works. Read a sentence, sing, or talk naturally.
How it works
Sound to picture
Your audio is turned into a spectrogram: 80 frequency bands, 100 times a second — the same input the model was trained on.
Whisper
OpenAI's open-source Whisper model (base size, 74 million parameters) listens to the spectrogram and predicts which language token comes first.
On your device
The model runs with ONNX Runtime in WebAssembly. Nothing is sent anywhere; after the first visit it works from your device's storage.
Built with OpenAI's Whisper (MIT) and Microsoft's ONNX Runtime Web (MIT), hosted on this site.