whatmodel.sh / tts / MetaVoice 1B v0.1

MetaVoice 1B v0.1

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

MetaVoice 1B v0.1 is an open-weights text-to-speech model from MetaVoice, released February 2024. Its published size is 1.2B parameters. Licence: apache-2.0.

Catalogue identity notes

MetaVoice-1B; https://huggingface.co/metavoiceio/metavoice-1B-v0.1. Submission inference details: Overwrote 30 second prompt constraint.. No weight commit supplied.

Model information sources

Release date ↗ · Weights licence ↗ · Parameter count ↗

Maker's figure for the base model; card also mentions a separate ~10Mn non-causal transformer, and uses multi-band diffusion + DeepFilterNet post-processing; whether 1.2B includes those is not stated.

19.8±3.1#114 of 118 text-to-speech modelsfrom 9 benchmark results across 5 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 MetaVoice 1B v0.1, out of 148 with a score.

#modelscorewhereRAMreleased
139Bark Suno25±3local~4.5GB2023-04
140SpeechT5 TTS Microsoft23±3local~0.7GB2023-02
141MetaVoice 1B v0.1 MetaVoice20±3local~9.7GB2024-02
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

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

$ what it's good at

other tests9 not in a task · 96% of its score
testresultvs the others in its rundifficultycountswhy
TTSDS2 listening test · speaker-similarity MOS (4 English domains)2.18 / 5 ✎▅·█▅▅·█▃▃▅·▅ beats 1/17easy · avg 4.03 / 5▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 64%at its level
SAP voice reconstruction · human BWS reconstruction preference795.7 ✎█▄█····▄··▄▄ beats 1/7avg 995▇▇▇▇ 11%typical
TTSDS2 listening test · naturalness MOS (4 English domains)2.49 / 5 ✎▅··▃▅··▅█▅██ beats 4/17medium · avg 3.30 / 5▇▇ 5.8%blunt · at its level
TTSDS2 multilingual benchmark · English overall (YouTube, 2025 round)86.3% ✎▂▂█·▂·█▆▂·▄▂ beats 8/17very easy · avg 89%▇▇ 5.6%blunt
CodecMOS-Accent VCTK voice clone · human speaker-similarity MOS (S-SPK-SIM)4.06 / 5 ✎▄▄█·▄▄▄▄▄·▄█ beats 5/11easy · avg 4.20 / 5▇ 3.4%blunt · at its level
CodecMOS-Accent VCTK voice clone · human naturalness MOS (S-NAT)3.07 / 5 ✎█··▄▄██▄·▄▄▄ beats 1/11easy · avg 3.99 / 5▇ 3.0%blunt · at its level
SAP voice reconstruction · TTSDS dual-reference distributional score79.2% ✎▄····▄▄▄▄▄·█ beats 0/7very easy · avg 86%▇ 2.4%blunt
TTSDS v1 · default four-factor score80.7% ✎▂▂····▃·▃▅▂█ beats 2/14very easy · avg 86%▇ 1.0%blunt
1 more test
testresultvs the others in its rundifficultycountswhy
SAP voice reconstruction · WER of reconstructed speech (Whisper turbo)43.3% ✎▄▄▄··█▄··▄·▄ beats 0/7avg 17%▇ 0.2%blunt · noisy

$ where to run it

Other ways to run it

  1. Run it as published · needs ~9.72 GB, NVIDIA only · from github.com/metavoiceio
    git clone https://github.com/metavoiceio/metavoice-src.git && cd metavoice-src && pip install -r requirements.txt && pip install torch==2.2.1 torchaudio==2.2.1 && pip install -e . && python -c "from fam.llm.fast_inference import TTS; tts = TTS(); tts.synthesise(text='This is a demo of text to speech by MetaVoice-1B, an open-source foundational audio model.', spk_ref_path='assets/bria.mp3')"

Official weights and documentation: huggingface.co/metavoiceio ↗

$ how the score was worked out · TTSDS2 listening test results… 70%, An Evaluation Framework for T… 13%, CodecMOS-Accent: listening-te… 6% and 2 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

sourceresultsshare of the scorelicence
TTSDS2 listening test results (ttsds/listening_test)2▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 70%MIT
An Evaluation Framework for TTS Voice Reconstruction (SAP dysarthric speakers; arXiv:2606.21343v1)3▇▇▇▇ 13%CC BY 4.0
CodecMOS-Accent: listening-test MOS of zero-shot TTS across English accents (arXiv:2603.14328v1)2▇▇ 6%CC BY 4.0
TTSDS2 multilingual benchmark results (ttsds/leaderboard_data)1▇▇ 6%MIT
TTSDS v1 published results1▇ 1%MIT
prior · pull toward the middle▇ 4%
weighted average= 19.8

$ 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