Open-source OmniVoice TTS ships voice cloning across 600+ languages at 40x realtime
abhishek__AI · x · 2026-10-20
k2-fsa released OmniVoice, an open-source (Apache-2.0) massively multilingual zero-shot TTS model that already has 15.1k GitHub stars, claiming the broadest language coverage among zero-shot TTS models.
- Supports 600+ languages with zero-shot voice cloning and voice design from text prompts
- Built on a novel diffusion language model-style architecture, running at 40x realtime, with a further 2.9x inference speedup via FlashInfer
- Ships with Python API, CLI tools, and training/evaluation scripts for direct deployment
Related event: Open-Source TTS Model OmniVoice Covers 600+ Languages at 40x Real-Time(2 posts)→
More from Multimodal
- Qwen 2.1 Shows Worse Subject-Scene Integration Than AIO v23 in Multi-Image Edit Tests — ShowerFluid9178 · 2026-10-20
- Edited a video entirely with AI: best half-day output yet, and it's 'scary' — aronkor · 2026-10-20
- Stanford's Halcyon: strong text-to-image without language pretraining — StanfordUniversity · 2026-10-20
- From one sentence to a finished trailer: a complete beginner's guide to AI filmmaking on a laptop — tetsuoai · 2026-10-20
- Asked GPT-6 to show polar bear vs. tiger size — it nailed the proportions — throwaway_anonymous7 · 2026-10-20
- Hyper-real candid morning couple photo: full Midjourney prompt shared — tisch_eins · 2026-10-20