whatmodel.sh / tts / dots.tts SOAR 2B
dots.tts SOAR 2B
↗ share
+ contribute
dots.tts SOAR 2B is an open-weights text-to-speech model from dots.tts, released June 2026. Its published size is 2B parameters. Licence: apache-2.0.
Catalogue identity notes
dots.tts SOAR 2B, not Pretrain or MeanFlow-distilled checkpoints.
Model information sources
Release date ↗ · Weights licence ↗ · Parameter count ↗
Maker's rounded system-level figure (semantic encoder + Qwen2.5-1.5B-initialised LLM + AR flow-matching head); not stated whether the frozen AudioVAE is counted.
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 dots.tts SOAR 2B, out of 148 with a score, then more from dots.tts.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 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 | ||
| 21 |  Eleven v3 | 65±1 | api $100.00/M chars | ||
| 22 |  Chirp 3: HD | 65±4 | api $30.00/M chars | ||
| more from dots.tts | |||||
| 9 | dots.tts MF 2B (MeanFlow, NFE 4) | 68±3 | local | ||
| 14 | dots.tts Base (Pretrain) 2B | 67±2 | comparison only | ||
see all text-to-speech models · find the ones that fit your machine
$ what it's good at
voice cloning▁·▁▁▁▁▁▁▇█▃▃ #2 of 37 · 30% of its score · 6 tests
Beats OpenAudio S1, Gemini 3.1 Flash TTS Preview, dots.tts MF 2B (MeanFlow, NFE 4) here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Seed-TTS test-zh · speaker similarity (ERes2Net SV) | 81.8% ✎ | █·····██···█ beats 2/3 | 6.9% |
| Seed-TTS test-en · speaker similarity (ERes2Net SV) | 79.2% ✎ | █····█··█··█ beats 2/3 | 6.8% |
| Seed-TTS test-en · speaker similarity (WavLM-large SV) lab | 77.1% ✎ | ▁·▂▂▂█▆▃▅··▂ beats 25/30 | 6.3% |
| Seed-TTS test-zh · speaker similarity (WavLM-large SV) lab | 81.0% ✎ | ▁·▂··▂·█▄·▃▄ beats 23/25 | 5.5% |
| Seed-TTS test-en · Whisper large-v3 WER lab | 1.3% ✎ | ▁···▂▂▂▂█▆▄▂ beats 32/35 | 2.4% |
| Seed-TTS test-zh · Paraformer CER lab | 0.9% ✎ | ▁·▂▁▁▁▂▃▅█·▂ beats 23/30 | 1.8% |
other tests8 not in a task · 69% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| CV3-Eval hard-en · WER +1 | 11.7% ✎ | █····█·█████ beats 0/6 | 21% |
| CV3-Eval hard-zh · CER +1 | 11.8% ✎ | ▄▄▄···▄··▄·█ beats 0/6 | 15% |
| CV3-Eval hard-en · DNSMOS (predicted quality) | 3.72 / 5 ✎ | █·▄·▄▄·▄···▄ beats 0/6 | 10% |
| CV3-Eval multilingual voice clone · 9-language mean error | 10.6% ✎ | ▄···▄··▄··██ beats 0/6 | 8.2% |
| CV3-Eval EN (zero-shot English) · WER lab | 4.5% ✎ | ▁·······▁▃█▇ beats 10/14 | 8.0% |
| CV3-Eval hard-en · speaker similarity (ERes2Net SV) | 76.0% ✎ | ▄·····▄▄·▄▄█ beats 5/6 | 4.4% |
| RVCBench LibriTTS clean voice cloning · ECAPA speaker similarity | 60.0% ✎ | ▃·····▂▂▂▃█▆ beats 14/15 | 3.3% |
| RVCBench LibriTTS clean voice cloning · Whisper-medium WER | 6.0% ✎ | ▂▂▂··▂·▃▃▃█▂ beats 8/15 | 0.0% |
$ where to run it
Other ways to run it
- Run it as published · needs ~5.16 GB, NVIDIA only · from github.com/studio-dots-ai
pip install dots.tts && dots.tts --model-name-or-path dots-studio/dots.tts-soar --text "Hello, this is a zero-shot voice cloning demonstration." --prompt-audio /path/to/reference.wav --prompt-text "The exact transcript of the reference audio." --output clone.wav
Official weights and documentation: huggingface.co/dots-studio ↗
$ how the score was worked out · Qwen-Audio-3.0-TTS technical… 56%, dots.tts Technical Report 40%, RVCBench: Benchmarking Robust… 3%
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 |
|---|---|---|
 Qwen-Audio-3.0-TTS technical report | 7 | 56% |
| dots.tts Technical Report · lab-reported | 7 | 40% |
| RVCBench: Benchmarking Robustness of Voice Cloning Across Modern Audio Generation Models (arXiv:2602.00443v2) | 2 | 3% |
| prior · pull toward the middle | 1% | |
| weighted average | = 65.2 |
$ 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