This speech model can even capture microexpressions, with millions of personality combinations. In the future, virtual hosts might be indistinguishable from real people.

StepAudio 2.5 Realtime is a real-time speech model capable of recognizing paralinguistic features such as tone, speech rate, pauses, and microexpressions. Through the API, it can integrate custom personalities, set traits, background stories, and language styles, with over ten thousand native personality options, allowing for millions of possible feature combinations. It comes with 5 built-in preset personalities, fine-tuned with RLHF, maintaining consistency even in complex role-playing scenarios, supporting both Chinese and English.