Particle.news
Download on the App Store
9 ARTICLES
·
2w ago

Amazon Launches Nova Sonic AI Voice Model for Real-Time Conversational Applications

The unified model integrates speech recognition and generation, offering developers advanced tools for natural interactions via Amazon Bedrock.

Image
Amazon has launched Nova Sonic, a new AI voice model built for faster, more accurate, and natural conversations.
Andy Jassy, chief executive officer of Amazon.com Inc., speaks during an unveiling event in New York, US, on Wednesday, Feb. 26, 2025. Amazon has rebooted Alexa with artificial intelligence, marking the biggest overhaul of the voice-activated assistant since its introduction over a decade ago. Photographer: Michael Nagle/Bloomberg via Getty Images
Amazon has unveiled its new human-like AI voice model, Nova Sonic.

Overview

  • Nova Sonic combines speech-to-text, language understanding, and text-to-speech into a single system, reducing development complexity and preserving conversational nuances.
  • The model captures tone, pacing, and emotional cues, enabling more natural and responsive interactions in real-time conversations.
  • Available now through Amazon Bedrock's bi-directional streaming API, it supports enterprise use cases like customer service, education, and entertainment.
  • Nova Sonic achieves industry-leading performance with a word error rate of 4.2% and latency of 1.09 seconds, making it 80% more cost-effective than competitors like OpenAI’s GPT-4o.
  • As part of Amazon’s broader AGI strategy, Nova Sonic is already powering features in Alexa+ and aims to expand into multimodal AI capabilities in the future.