Speech-to-Text (STT)

🚧
Beta: Speech-to-Text is in beta. Models, messages and limits may change before general availability.

Stream audio to Krisp over a WebSocket and get transcripts back in realtime. Typical uses are live captions, meeting transcription and voice agents.

The STT endpoint serves two models. They accept different handshake fields and send different server messages, so each one has its own page:

Model	Built for	Main features	Page
`krisp-realtime-stt-humans-v1`	Human conversations: meetings, calls, live captions	Speaker diarization, PII redaction, custom vocabulary	krisp-realtime-stt-humans-v1
`krisp-realtime-stt-agents-v1`	Voice AI agents	Fast partial results, end-of-turn prediction (`turn` messages), force finalization, custom vocabulary	krisp-realtime-stt-agents-v1

Try it

Paste an API key from the Developers Portal, choose a model, press Start and speak. The demo creates a short-lived session token from your key and streams your microphone to wss://streaming.krisp.ai/stt. Sessions count toward your usage.

Common to both models

WebSocket URL: wss://streaming.krisp.ai/stt
Auth: a temporary session token, sent as Authorization: api-key <session-token>, or as the authorization query parameter if headers can't be set. Do not send your master token on the WebSocket. See Authentication.
Handshake: the first message must be a JSON handshake that selects the model. It must arrive within 15 seconds. Wait for the ready message before sending audio.
Audio: raw bytes in binary WebSocket messages, 16,000 Hz, mono, pcm_s16le or opus. Recommended chunk duration is 80 ms or a multiple of 80 ms. Each message can be at most 65,536 bytes.
Stream end: send {"type":"end_of_audio"} after the last audio message, then read messages until end. The server closes the socket.
Errors: see the error table on each model page and the common errors.

Changelog

2026-10-01: Split into one page per model: krisp-realtime-stt-humans-v1 and krisp-realtime-stt-agents-v1.
2026-10-01: Authentication now uses a session token; see Authentication.
2026-10-01: Public model list limited to krisp-realtime-stt-humans-v1 and krisp-realtime-stt-agents-v1.
2026-10-01: Page created: client hello, server messages, supported models.