Xiaomi launches MiMo-V2.5-ASR voice model with top-tier recognition in tough scenarios

Xiaomi has launched the MiMo-V2.5-ASR voice model, designed as a full-chain speech base. It claims top-level performance in tough scenarios like dialects, mixed languages, noisy environments, and multi-speaker conversations. The model supports Chinese dialects such as Cantonese, Wu, Minnan, and Sichuanese, and handles code-switching between Chinese and English seamlessly. It also recognizes songs, works well in high noise, and transcribes multiple speakers accurately. Plus, it identifies ancient poems, technical terms, and names, and outputs punctuation directly for ready-to-use text. The API is now available, priced per hour of audio: 0.5 yuan in China and $0.074 overseas.