whatmodel.sh / tts / LEMAS-TTS (w/ prosody, multilingual_prosody)
LEMAS-TTS (w/ prosody, multilingual_prosody)
↗ share
+ contribute
LEMAS-TTS (w/ prosody, multilingual_prosody) is an open-weights text-to-speech model from LEMAS Project, released January 2026. Licence: cc-by-4.0.
Catalogue identity notes
LEMAS-Project/LEMAS-TTS @ 386ba35 (cc-by-4.0, created 2026-01-01) ships two TTS checkpoints; this entry is pretrained_models/ckpts/multilingual_prosody/multilingual_prosody.safetensors = the report's 'LEMAS-TTS w/ prosody' row (arXiv:2601.04233v1 Table 4; 'We therefore release both variants'). Mapping evidence: tts_multilingual.py 'if model_name.endswith("grl"): use_prosody_encoder = False' and speech_edit_multilingual.sh '# Main example: multilingual_grl, no prosody encoder' vs the multilingual_prosody example with --enable_prosody_encoder. Proposed so the lone OpenAudio-S1-Mini lemas-eval-9lang-wer run counts (3 models).
Model information sources
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 LEMAS-TTS (w/ prosody, multilingual_prosody), out of 148 with a score, then more from LEMAS Project.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 72 |  Gepard 1.0 | 53±4 | comparison only | ||
| 73 |  Kokoro 82M v1.0 | 52±3 | api $0.62/M chars +local | ||
| 74 | LEMAS-TTS (w/ prosody, multilingual_prosody) | 52±6 | local | ||
| 75 |  Neuphonic neu_hq | 52±3 | comparison only | ||
| 76 |  Spark-TTS 0.5B | 52±3 | local | ||
| more from LEMAS Project | |||||
| 84 | LEMAS-TTS (w/o prosody, multilingual_grl) | 50±6 | local | ||
see all text-to-speech models · find the ones that fit your machine
$ what it's good at
other tests1 not in a task · 85% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| LEMAS multilingual TTS eval · 9-language mean WER lab | 4.8% ✎ | █········█·█ beats 2/2 | 85% |
$ where to run it
Other ways to run it
- Run it as published · needs ~1.45 GB · from github.com/LEMAS-Project (official)
git clone https://github.com/LEMAS-Project/LEMAS-TTS.git && cd LEMAS-TTS && pip install -r requirements.txt && hf download LEMAS-Project/LEMAS-TTS --include "pretrained_models/*" --local-dir . && PYTHONPATH=$PWD python -m lemas_tts.scripts.tts_multilingual --model multilingual_prosody --ckpt_file pretrained_models/ckpts/multilingual_prosody/multilingual_prosody.safetensors --vocab_file pretrained_models/data/multilingual_prosody/vocab.txt --frontend phone --use_ema --enable_prosody_encoder --ref_audio pretrained_models/demos/lemas_tts_test/es.wav --ref_text "Te voy a dar un tip que le copié a John Rockefeller, uno de los empresarios más picudos de la historia. " --text "I want to give you a tip I copied from John Rockefeller." --output_wave out.wav --nfe_step 64 --cfg_strength 5.0 --sway_sampling_coef 3.0 --ref_ratio 1.0 --separate_langs --speed 1.0 --seed -1
Official weights and documentation: huggingface.co/LEMAS-Project ↗
$ how the score was worked out · LEMAS multilingual audio suit… 85%
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 |
|---|---|---|
 LEMAS multilingual audio suite report · lab-reported | 1 | 85% |
| prior · pull toward the middle | 15% | |
| weighted average | = 52.2 |
$ 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