whatmodel.sh / tts / OmniVoice
OmniVoice
↗ share
+ contribute
OmniVoice is an open-weights text-to-speech model from k2-fsa (Xiaomi), released March 2026. Its published size is 0.61B parameters. Licence: cc-by-nc-4.0.
Catalogue identity notes
k2-fsa/OmniVoice (Qwen3-0.6B based; 612,577,288 parameters per HF safetensors metadata). Code Apache-2.0; README: 'The pre-trained model is licensed under the CC-BY-NC due to constraints from its training data'.
Score within the Vapi Humanness · Bradley–Terry rating, 2026-09-25 snapshot comparison set only.
$ how it compares
The text-to-speech models ranked either side of OmniVoice, out of 148 with a score, then more from k2-fsa (Xiaomi).
| # | model | score | where | ||
|---|---|---|---|---|---|
| 16 |  LongCat-AudioDiT-3.5B | 66±2 | local | ||
| 17 | LongCat-AudioDiT-1B | 66±4 | local | ||
| 18 | OmniVoice | 66±2 | local | ||
| 19 |  MiniMax Speech 2.8 HD | 66±2 | api $100.00/M chars | ||
| 20 |  dots.tts SOAR 2B | 65±2 | local | ||
| more from k2-fsa (Xiaomi) | |||||
| 42 | ZipVoice | 62±3 | local | ||
see all text-to-speech models · find the ones that fit your machine
$ what it's good at
voice cloning▁·▁▁▁▁▁▁▇█▃▃ #10 of 37 · 67% of its score · 8 tests
Beats Gemini 3.1 Flash TTS Preview here, though they rank higher overall. Trails VoxCPM2, Higgs Audio V3 TTS, dots.tts SOAR 2B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Seed-TTS test-en · speaker similarity (WavLM-large SV) lab +1 | 74.1% ✎ | ▁·▂▂▂█▆▃▅··▃ beats 29/30 | 18% |
| Seed-TTS-Eval · test-en/test-zh macro mean error | 1.2% ✎ | ▄··▄▄····█▄▄ beats 5/6 | 13% |
| Seed-TTS test-en WER (transcriber not stated) | 1.6% ✎ | ▃··········█ beats 1/3 | 10% |
| Seed-TTS test-zh · speaker similarity (WavLM-large SV) lab | 77.7% ✎ | ▁·▂··▂·█▄·▃▄ beats 18/25 | 8.4% |
| nsanku TTS Benchmark · 7 Ghanaian languages mean CER (GhanaNLP) | 14.1% ✎ | █··········▄ beats 2/2 | 5.9% |
| Seed-TTS test-en · Whisper large-v3 WER lab | 1.6% ✎ | ▁···▂▂▂▂█▆▄▂ beats 26/35 | 4.3% |
| Seed-TTS test-en · UTMOS (predicted MOS) | 2.63 / 5 ✎ | █·····██···█ beats 1/3 | 4.2% |
| Seed-TTS test-zh · Paraformer CER lab | 0.8% ✎ | ▁·▂▁▁▁▂▃▅█·▂ beats 27/30 | 2.8% |
other tests6 not in a task · 32% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Wayu Thai TTS eval · 1,531-item Challenge-Set keyword accuracy | 72.8% ✎ | █···█······█ beats 1/2 | 8.8% |
| Wayu Thai TTS eval · English CER on 500 held-out LibriTTS lines (Whisper Large V3) | 0.9% ✎ | █······█···█ beats 1/2 | 8.1% |
| Wayu Thai TTS eval · 500-utterance Thai CER (Typhoon Whisper Large V3) | 4.6% ✎ | █······█···█ beats 0/2 | 7.5% |
| MiniMax multilingual test set · 23-language mean error | 3.0% ✎ | ▃·▃·▃·····▃█ beats 5/6 | 2.7% |
| Arabic TTS Arena · community blind A/B rating, 2026-09-24 snapshot | 1049.4 ✎ | ▂▂··▃·█▅···▂ beats 5/12 | 2.6% |
| CV3-Eval multilingual voice clone · 9-language mean error | 4.9% ✎ | ▃·····▃···▅█ beats 4/6 | 2.0% |
$ where to run it
Other ways to run it
- Run it as published · needs ~3.26 GB · from github.com/k2-fsa (official)
pip install omnivoice && omnivoice-infer --model k2-fsa/OmniVoice --text "This is a test for text to speech." --output hello.wav
Official weights and documentation: huggingface.co/k2-fsa ↗
$ how the score was worked out · OmniVoice: Towards Omnilingua… 26%, Building and Evaluating Fixed… 24%, nineninesix/gepard-1.0 model… 23% and 3 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 OmniVoice: Towards Omnilingual Zero-Shot Text-to-Speech with Diffusion Language Models (arXiv:2604.00688v3) · lab-reported | 4 | 26% |
| Building and Evaluating Fixed-Voice Thai TTS from Synthetic Speech (Wayu Research / Paxa Labs / Typhoon; arXiv:2609.03502v1) | 3 | 24% |
 nineninesix/gepard-1.0 model card (Seed-TTS-eval benchmark table) | 3 | 23% |
| Higgs Audio v3 comparison tables in SGLang-Omni cookbook (Boson AI contribution) | 3 | 17% |
| nsanku TTS Benchmark (GhanaNLP community Space) | 1 | 6% |
| Arabic TTS Arena (Navid AI) · leaderboard snapshot 2026-09-24 | 1 | 3% |
| prior · pull toward the middle | 2% | |
| weighted average | = 65.7 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json