MiMo-V2-TTS

Aibase reported on March 19, 2026 that Xiaomi officially released its self-developed large-scale speech synthesis model MiMo-V2-TTS, describing its basis in Xiaomi's self-developed Audio Tokenizer and a multi-codebook speech-text joint modeling architecture, pre-trained on hundreds of millions of hours of speech data. The third-party report highlights cross-regional adaptability with support for multiple dialects (Northeastern Mandarin, Sichuanese, Henanese, Cantonese, Taiwanese accent) and character-based performance, as well as intelligent recognition of text signals such as punctuation, intonation words, and emphasis markers with

On March 19, 2026, Xiaomi officially launched MiMo-V2-TTS, a self-developed speech synthesis large model positioned as a versatile voice actor capable of speaking, performing, and singing. According to the article, the model is built on Xiaomi's proprietary Audio Tokenizer with a multi-codebook speech-text joint modeli The coverage highlights specific capability pillars: an "Emotion Master" mode that shifts tone and subtly changes emotion across a single sentence; a "Cross-over Singer" mode that handles pitch, rhythm, and expressive vocal techniques for singing; and a "Dialect Expert" mode that supports Northeastern, Sichuan, Henan,

The AIBase report confirms the March 19, 2026 release of MiMo-V2-TTS as part of Xiaomi's three-model MiMo-V2 launch, describing it as the speech model giving Agents "warm" expression with fine-grained emotional control. The article places V2-TTS within Xiaomi's "Agent stack" alongside the >1T-parameter MiMo-V2-Pro and The same coverage notes that MiMo-V2-Pro ranks eighth globally on Artificial Analysis' comprehensive intelligence ranking, and that Lei Jun publicly committed more than 16 billion yuan to AI R&D and capital investment in 2026. Pro and Omni APIs opened at launch; Pro input pricing was quoted at $1 per million tokens wit