Amazon Nova Sonic
Single speech-to-speech model unifying understanding and generation, with a new bidirectional streaming API in Bedrock.
Replaces the transcribe-LLM-synthesize pipeline with one model that adapts prosody and tone to context, supports function calling and RAG, and watermarks output; launched in US East with English voices and accents.
- Date
- Tuesday, 8 April 2025
- Lab
- Amazon
- Kind
- model
- Access
- closed API
AWS gave no numeric latency figure in the announcement.
Sources
This record was checked against its sources on 6 October 2026. How we check