Amazon Debuts Nova Sonic, Its Most Advanced AI Voice Model

The unified voice AI, now live via Amazon Bedrock, sets new benchmarks in speed, cost-efficiency, and conversational naturalness.

Nova Sonic combines speech recognition, language understanding, and speech generation into a single model for seamless, human-like voice interactions.
The model is available through Amazon Bedrock’s bi-directional streaming API, enabling developers to integrate it into third-party applications.
Nova Sonic outperforms competitors with a 4.2% word error rate, 1.09-second latency, and nearly 80% lower cost compared to OpenAI’s GPT-4o.
It excels in multilingual and noisy environments, preserving tone, cadence, and context for natural conversation even with interruptions or accents.
Already powering Amazon's upgraded Alexa+ assistant, Nova Sonic is part of Amazon’s broader roadmap toward artificial general intelligence (AGI).