Xiaomi Releases MiMo-V2.5 Voice Models: Flexible Sound Control via Natural Language

Xiaomi today released the MiMo-V2.5-TTS Series and MiMo-V2.5-ASR voice models, covering two core capabilities: speech recognition and synthesis. These models allow voice input and output to be flexibly controlled via natural language, targeting the Agent era to enhance the intelligence and creativity of voice interactions.

The MiMo-V2.5-TTS Series includes three models, now available on the Xiaomi MiMo Open Platform for a limited free period. The models support style instruction following, director-script-level structured input, and inline audio tags, enabling precise control over sound details like emotion and speed. Even plain text can be rendered with natural rhythm and emotion.

As an auditory base, MiMo-V2.5-ASR performs leadingly in complex scenarios such as Chinese-English bilingualism, dialects, and strong noise. It also open-sources model weights and code for developers to secondary develop.