NEW YORK, April 8: On Tuesday, Amazon introduced Nova Sonic, a new generative AI voice model that produces natural-sounding speech and excels in speed, speech recognition, and conversational quality, competing with models from OpenAI and Google. Nova Sonic is available through Amazon's Bedrock platform, with a new bi-directional streaming API. Amazon claims it is 80% more cost-efficient than OpenAI’s GPT-4o.

Nova Sonic is already integrated into Alexa+, Amazon's upgraded voice assistant, and leverages Amazon's expertise in “large orchestration systems” for routing user requests to the right APIs. It is designed to handle two-way conversations naturally, considering pauses and interruptions, and offers a text transcript of user speech.

The model is highly accurate, with a 4.2% word error rate on the Multilingual LibriSpeech benchmark across multiple languages. It also outperformed OpenAI's GPT-4o by 46.7% in multi-participant interactions and has an impressive speed with a latency of 1.09 seconds, faster than GPT-4o’s 1.18 seconds.

Nova Sonic is part of Amazon’s broader strategy to develop Artificial General Intelligence (AGI). The company plans to release more AI models in the future, including ones that process images, video, and other sensory data.