Ant Group released its Ming-3.0 speech model family on October 20, making five models available across speech synthesis, audio understanding and conversation, and speech recognition. Developers can use individual models or combine them into voice-based applications.
The text-to-speech models support voice cloning from reference recordings and voice design through natural-language descriptions. Users can also direct delivery by specifying emotion, emphasis and pacing. The audio models handle transcription, summarization and analysis of long recordings while retaining information about speakers, emotion and background sounds. Ant said the models already support services including its Afu health app, Alipay search and internal office tools. [IT Home, in Chinese]