GitLab Transcend - Intelligent orchestration
Mini toolsOmniVoice
By Xiaomi
OmniVoice is a state-of-the-art multilingual TTS model from the Xiaomi Next-gen Kaldi team, built for zero-shot speech generation across more than 600 languages. Public project materials describe it as using a novel diffusion language model-style architecture over discrete audio tokens, with support for three main modes: voice cloning from reference audio, voice design from speaker attributes, and automatic voice generation without a reference clip. The project also highlights very fast inference, with reported real-time factors as low as 0.025, and an Apache 2.0 open-source release.
New
Multimodal
Gen 3
Overview
OmniVoice is a multilingual zero-shot text-to-speech model built for voice cloning, voice design, and general speech synthesis at massive language scale. It supports more than 600 languages, uses a diffusion language model-style architecture, and is positioned for high-quality speech generation with fast inference.
About Xiaomi
At Xiaomi, we believe technology’s the true power lies in its ability to understand and enhance the human experience. This year, with the upgraded Xiaomi HyperOS 2 and the newly launched Xiaomi HyperAI, we are redefining connection.
View Company ProfileTools using OmniVoice
No tools found for this model yet.
Last updated: April 7, 2026
