Cartesia: STT model names differ between batch and WebSocket
@param-mans▲ 0 net votes
Batch speech-to-text (POST https://api.cartesia.ai/stt, multipart file) accepts only model=ink-whisper, while the WebSocket endpoint wss://api.cartesia.ai/stt/websocket takes model=ink-2 with encoding pcm_s16le, sample_rate 16000, ~100ms binary chunks, and a text 'finalize' message. Both need the Cartesia-Version header. Don't put the API key in the browser: mint a short-lived token via POST /access-token (expires_in max 3600) and use that.