whatmodel.sh / tts / MMS TTS Turkish

MMS TTS Turkish

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

MMS TTS Turkish is an open-weights text-to-speech model from Meta, released September 2023. Its published size is 0.036B parameters. Licence: cc-by-nc-4.0.

Catalogue identity notes

Meta's Massively Multilingual Speech VITS checkpoint for Turkish, facebook/mms-tts-tur (pinned 7e36447, CC BY-NC 4.0). The Antalia re-run loads VitsModel.from_pretrained('facebook/mms-tts-tur') (harness/systems/run_mms.py).

Model information sources

Release date ↗ · Weights licence ↗

36M as printed (HF safetensors metadata 36,285,552).

18.0±3.9#116 of 118 text-to-speech modelsfrom 1 benchmark result across 1 source* thinly tested, treat the rank as provisional

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 MMS TTS Turkish, out of 148 with a score, then more from Meta.

#modelscorewhereRAMreleased
142Tacotron 2 + HiFi-GAN (VCTK, VITS study) VITS study (public implementations)19±6comparison only2021-06
143YourTTS Coqui / Edresson Casanova et al.18±3local~0.4GB2021-12
144MMS TTS Turkish Meta18±4local~0.2GB2023-09
145Antalia 1 Patientdesk.ai17±3comparison only2026-09
146Amphion NaturalSpeech 2 (LibriTTS) Amphion (CUHK-Shenzhen)14±3local~1.7GB2023-12
more from Meta
85MMS TTS Amharic Meta50±16local~0.2GB2023-05

see all text-to-speech models · find the ones that fit your machine

$ what it's good at

other tests1 not in a task · 94% of its score
testresultvs the others in its rundifficultycountswhy
Freya-TR-Eval Turkish TTS intelligibility (WER, faster-whisper large-v3, 8 kHz band)6.4% ✎▄··▄··▄···▂█ beats 2/10avg 0.3%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 94%typical

$ where to run it

Other ways to run it

  1. Run it as published · needs ~0.15 GB · from huggingface.co/facebook
    uv run --python 3.11 --with transformers --with torch --with scipy python -c "from transformers import VitsModel, AutoTokenizer; import torch, scipy; model = VitsModel.from_pretrained('facebook/mms-tts-tur'); tokenizer = AutoTokenizer.from_pretrained('facebook/mms-tts-tur'); inputs = tokenizer('Merhaba, nasılsınız?', return_tensors='pt'); torch.set_grad_enabled(False); output = model(**inputs).waveform; scipy.io.wavfile.write('out.wav', rate=model.config.sampling_rate, data=output.squeeze().numpy())"

Weights and documentation: huggingface.co/facebook ↗

$ how the score was worked out · Antalia-2 Mini model card (cl… 94%

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

sourceresultsshare of the scorelicence
Antalia-2 Mini model card (cloud0day3/antalia-mini), public Freya-TR-Eval re-run table1▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 94%Apache-2.0
prior · pull toward the middle▇▇ 6%
weighted average= 18.0

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

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