Try Ink 2, which has improved transcription accuracy and turn
detection for voice agents.
ink-whisper
Ink Whisper is our most affordable speech-to-text model. It’s more accurate than baseline Whisper and has lower latency.
Compared to baseline Whisper, Ink Whisper also:
- Uses dynamic chunking to handle variable-length audio chunks and interruptions.
- Reliably transcribes speech with background noise.
- Accurately transcribes audio with telephony artifacts, accents, and disfluencies.
- Accurately transcribes proper nouns and domain-specific terminology.
To learn how to use the Ink STT models, see Compare STT endpoints.