Cartesia builds real-time speech models and exposes them through streaming APIs for voice agents. The line-up covers both directions of a conversation: Sonic generates speech from text, Ink transcribes speech to text, and Line is a platform for assembling the two into a deployable voice agent.
Sonic targets conversational latency, with time-to-first-audio under 90 ms, and supports more than 40 languages and accents. It clones a voice from about ten seconds of audio, interprets emotional subtext in the text to shape delivery, keeps pacing consistent, accepts non-verbal cues such as laughter inline in the transcript, and honors pronunciation dictionaries for proper nouns and domain terms.
Ink is a streaming speech-to-text model with built-in turn detection that marks where a speaker starts and stops, so an agent can respond without a separate endpointing step. Line wraps the models with audio orchestration and observability so you write the agent logic rather than the media plumbing. Both models are reachable over WebSocket and REST, with Python and JavaScript SDKs and a browser playground.
Features
- Sonic text-to-speech: sub-90 ms latency, 40+ languages and accents, emotion-aware delivery
- Instant voice cloning: a new voice from roughly ten seconds of reference audio
- Speech controls: speed, inline non-verbal expressions, and custom pronunciation dictionaries
- Localization: the same voice and tone carried across languages
- Ink speech-to-text: streaming transcription with native turn detection
- Line voice agent platform: orchestration and observability around Sonic and Ink
- Streaming APIs: WebSocket and REST endpoints with Python and JavaScript SDKs
- Playground: browser-based testing of models and API key management
