MiMo-V2-TTS is a large-scale speech synthesis model developed by Xiaomi as part of its second-generation multimodal AI suite. Launched in March 2026 alongside the MiMo-V2-Pro and MiMo-V2-Omni models, it is designed to provide AI agents with highly expressive and human-like vocal capabilities. The model serves as the primary audio generation engine for Xiaomi's "Human x Car x Home" ecosystem, facilitating natural voice interactions across consumer electronics, smart home devices, and electric vehicles.
The model's architecture is built on a proprietary Audio Tokenizer that employs multi-codebook joint modeling to effectively align speech and text data. MiMo-V2-TTS was trained on a massive dataset exceeding 100 million hours of audio, which allows it to capture a wide spectrum of vocal nuances. To enhance output quality, the model utilizes multi-dimensional reinforcement learning to refine its prosody, ensuring that generated speech sounds natural rather than robotic.
Key capabilities of MiMo-V2-TTS include fine-grained emotional control and the ability to adjust speaking styles and tones to suit different contexts. It supports multiple Chinese dialects and can transition between standard conversational speech and more creative outputs, such as singing. This focus on "warmth" and emotional intelligence is intended to make machine interactions feel more personified and accessible for end-users.