Streaming speech to text

Transcribe live audio over a WebSocket.

/v1/speech-to-text/stream

Open a WebSocket with your key in the x-api-key header (or ?api_key= where headers cannot be set), send a start message, then raw 16-bit little-endian mono PCM in binary frames. Each time the speaker pauses, the utterance is transcribed and sent back.

Direction	Message
you →	`{"type": "start", "language_code"?: "en-IN", "model_id"?, "sample_rate"?: 16000, "partials"?: false}`
← server	`{"type": "ready", "request_id"}`
you →	binary PCM frames…
← server	`{"type": "partial", "text"}`
← server	`{"type": "final", "text", "language_code", "start", "end", "audio_seconds"}`
you →	`{"type": "end"}`
← server	`{"type": "done", "audio_seconds"}`
← server	`{"type": "error", "code", "message"}`

Partials (the utterance so far, while the person is still speaking) are sent only with "partials": true. Audio is charged as it arrives; when credits run out the server sends an INSUFFICIENT_CREDITS error and closes.

from lipivo import AsyncLipivo   # pip install "lipivo[streaming]"

async with AsyncLipivo() as client:
    async for event in client.speech_to_text_stream(microphone_chunks(), language_code="en-IN"):
        if event["type"] == "final":
            print(event["text"])

← Speech to text Models, languages and voices →