Cartesia Ink vs Google Cloud Speech-to-Text

Cartesia Ink

Ink 2 streams transcripts with turn detection built into the model, documented by an AsyncAPI file with typed ranges and structured errors. The terms, last revised 14 June 2024, let Cartesia train on inputs and outputs unless the customer opts out, and zero data retention is an Enterprise setting. Ink 2 has no batch endpoint and no diarisation.

Google Cloud Speech-to-Text

Audio isn't stored or used for training unless the project opts in to data logging. No release note since 2025-11-13.

Cartesia Ink

1. Use wss://api.cartesia.ai/stt/turns/websocket with model=ink-2, encoding, sample_rate and cartesia_version=2026-08-14. All four are required.
2. Send raw mono audio in chunks of about 100 ms at the speed it was spoken. Pushing a whole file into the socket can return an internal server error.
3. Read the final text from turn.end only. transcript is cumulative within a turn, so joining turn.update events duplicates text.
4. Send {"type": "close"} after the last audio and keep reading until the server closes the socket, or the buffered tail is lost.
5. Check encoding and sample_rate against the source before sending. The docs say the server might not return an error when they are wrong.

Google Cloud Speech-to-Text

1. Call Chirp 3 on the us or eu endpoint. It isn't listed for the global location
2. Reopen streams before the 5-minute limit, or use BatchRecognize for recordings
3. Downmix stereo unless you need channel labels, since each channel is billed
4. Set dynamic batch on offline jobs to cut the price from $0.016 to $0.003 a minute
5. Back off on RESOURCE_EXHAUSTED. The Speech docs don't give a retry interval