Speech-to-text API. Record from the microphone below, or integrate with
POST /v1/transcribe. Full reference at /docs.
The key is kept in this browser's local storage and sent only to this server. It is never embedded in the page source.
Idle.
Quiet segments are discarded in the browser and never uploaded. The server repeats that check on measured loudness, because the model invents sentences when handed silence. Continuous mode restarts the recorder each segment so every upload is a complete, decodable file.