A new speech-to-speech AI model from Amazon, called Nova Sonic, unifies speech recognition and generation to deliver more natural voice interactions — part of the Seattle tech giant’s broader effort to develop human-like intelligence in competition with Google, OpenAI, and others.