The new text-to-speech model from Xiaomi’s Kaldi team offers high-quality voice cloning across hundreds of languages, demonstrating superior performance in low-resource linguistic environments.
Xiaomi AI Lab’s new-generation Kaldi team officially open-sourced its OmniVoice text-to-speech (TTS) model on May 8, 2026, marking a significant advancement in multilingual voice cloning and customizable speech generation.
The company announced via its official WeChat account that OmniVoice delivers high-quality speech synthesis across hundreds of languages, performing strongly in both Chinese and English scenarios, and often surpassing existing commercial systems in multilingual tasks.
A primary feature of OmniVoice is its robust support for low-resource languages, enabling speech generation in
