whatmodel.sh / tts / FireRedTTS3 (base)
FireRedTTS3 (base)
↗ share
+ contribute
FireRedTTS3 (base) is an open-weights text-to-speech model from FireRedTeam (Xiaohongshu), released August 2026. Its published size is 2.12B parameters. Licence: apache-2.0.
Catalogue identity notes
FireRedTeam/FireRedTTS3 run in its base (zero-shot cloning) mode. Distinct from FireRedTTS (v1), FireRedTTS-1S and FireRedTTS-2.
Model information sources
Base TTS model only: OUR COUNT from the safetensors header of the official fireredtts3_base/model.safetensors (pinned revision; 2,120,632,897 F32). The separate RedAE audio codec (redae/model.safetensors, 943.8M) is not included; the card states no parameter count. Observed 2026-10-07.
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 FireRedTTS3 (base), out of 148 with a score, then more from FireRedTeam (Xiaohongshu).
| # | model | score | where | ||
|---|---|---|---|---|---|
| 10 |  Raon-OpenTTS-1B | 68±2 | comparison only | ||
| 11 |  Grok TTS | 68±3 | api $15.00/M chars | ||
| 12 | FireRedTTS3 (base) | 67±3 | local | ||
| 13 |  Simba 3.2 | 67±3 | api $6.00/M chars | ||
| 14 |  dots.tts Base (Pretrain) 2B | 67±2 | comparison only | ||
| more from FireRedTeam (Xiaohongshu) | |||||
| 90 | FireRedTTS (v1) | 48±3 | local | ||
see all text-to-speech models · find the ones that fit your machine
$ what it's good at
voice cloning▁·▁▁▁▁▁▁▇█▃▃ #6 of 37 · 97% of its score · 5 tests
Beats Gemini 3.1 Flash TTS Preview here, though they rank higher overall. Trails dots.tts SOAR 2B, LongCat-AudioDiT-3.5B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Seed-TTS test-en · speaker similarity (WavLM-large SV) | 77.2% ✎ | █·█·██·▄·▄·▄ beats 10/10 | 29% |
| Seed-TTS test-zh · speaker similarity (WavLM-large SV) | 80.8% ✎ | ▃··▃█▅▅··▃·▃ beats 10/10 | 28% |
| MiniMax multilingual test set · 21-language mean error (UltraEval-Audio common set) | 5.4% ✎ | █··█·····█·█ beats 3/3 | 18% |
| Seed-TTS test-en · Whisper large-v3 WER | 1.8% ✎ | ▃··▃▅··▅·█▃▅ beats 7/11 | 12% |
| Seed-TTS test-zh · Paraformer CER | 0.9% ✎ | ▂··▄··▂▂█·▄▂ beats 9/11 | 11% |
$ where to run it
Other ways to run it
- Run it as published · needs ~12.3 GB, NVIDIA only · from huggingface.co/FireRedTeam (official)
git clone https://github.com/FireRedTeam/FireRedTTS3.git && cd FireRedTTS3 && pip install -r requirements.txt && pip install "huggingface_hub[cli]" && hf download FireRedTeam/FireRedTTS3 --local-dir pretrained_models/ && curl -L -o fireredtts3/utils/llm_tn/models/lid.176.ftz https://dl.fbaipublicfiles.com/fasttext/supervised-models/lid.176.ftz && python -c "import torch; import torchaudio; from fireredtts3.core import FireRedTTS3; tts = FireRedTTS3('pretrained_models', use_wetext=True, use_llm_tn=False); prompt_audio, prompt_audio_sr = torchaudio.load('prompt.wav'); gen_audio, gen_audio_sr = tts.generate(language=None, prompt_text='<prompt audio text>', prompt_audio=prompt_audio, prompt_audio_sr=prompt_audio_sr, text='今天天气很好，我们一起去公园散步吧。', do_tn=True); torchaudio.save('gen.wav', gen_audio.cpu(), gen_audio_sr)"
Official weights and documentation: huggingface.co/FireRedTeam ↗
$ how the score was worked out · UltraEval-Audio (OpenBMB) · T… 97%
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 |
|---|---|---|
 UltraEval-Audio (OpenBMB) · TTS replication results | 5 | 97% |
| prior · pull toward the middle | 3% | |
| weighted average | = 67.2 |
$ 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