Amazon Nova Sonic

Nova Sonic provides a single, full-duplex session that listens and talks at the same time. Incoming audio is transcribed as it arrives, the model begins reasoning on partial text, and synthesized speech is emitted continuously, so the assistant can reply mid-utterance, accept barge-ins, and course-correct without awkward pauses. The API exposes incremental transcripts, token streams, and audio frames, letting you plug in retrieval and tool calls while the conversation is still unfolding. Developers typically connect over WebRTC or websockets and shape outputs with controls for voice, style, and pacing. Because it sits within the broader Nova stack, Sonic brings production features like authentication, usage controls, and observability, and it drops cleanly into RAG and agent frameworks. Use it when you need lifelike voice interfaces with low-latency turn-taking—customer support, in-app copilots, and hands-free experiences in vehicles or devices.