whatmodel.sh / tts / OpenVoice V2

OpenVoice V2

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

OpenVoice V2 is an open-weights text-to-speech model from MyShell, released April 2024. Licence: mit.

Catalogue identity notes

OpenVoice V2; https://huggingface.co/myshell-ai/OpenVoiceV2. Submission inference details: not supplied. No weight commit supplied.

Model information sources

Release date ↗ · Weights licence ↗

25.7±3.0#111 of 118 text-to-speech modelsfrom 7 benchmark results across 4 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 OpenVoice V2, out of 148 with a score, then more from MyShell.

#modelscorewhereRAMreleased
136HierSpeech++ (LibriTTS-960) Korea University (Sang-Hoon Lee et al.)28±5local~2.1GB2023-11
137Piper (tr_TR dfki medium) Open Home Foundation (Piper)26±4local~0.1GB2023-06
138OpenVoice V2 MyShell26±3local~0.8GB2024-04
139Bark Suno25±3local~4.5GB2023-04
140SpeechT5 TTS Microsoft23±3local~0.7GB2023-02
more from MyShell
79OpenVoice V1 MyShell51±14local~0.3GB2024-01
103MeloTTS English MyShell44±5local~0.7GB2024-02

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

$ what it's good at

other tests7 not in a task · 96% of its score
testresultvs the others in its rundifficultycountswhy
TTSDS2 listening test · speaker-similarity MOS (4 English domains)2.59 / 5 ✎▅·█▅▅·█▃▃▅·▅ beats 3/17easy · avg 4.03 / 5▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 67%at its level
RVCBench LibriTTS clean voice cloning · ECAPA speaker similarity24.0% ✎▃·····▂▂▂▃█▆ beats 1/15medium · avg 46%▇▇▇▇ 10%blunt · at its level
TTSDS2 multilingual benchmark · Chinese overall (YouTube, 2025 round)82.0% ✎▄▄·····▄··▄█ beats 1/5very easy · avg 88%▇▇▇ 7.6%blunt
TTSDS2 listening test · naturalness MOS (4 English domains)2.92 / 5 ✎▅··▃▅··▅█▅██ beats 7/17medium · avg 3.30 / 5▇▇ 5.7%blunt · at its level
TTSDS2 multilingual benchmark · English overall (YouTube, 2025 round)82.5% ✎▂▂█·▂·█▆▂·▄▂ beats 4/17very easy · avg 89%▇▇ 5.0%blunt
TTSDS v1 · default four-factor score86.2% ✎▂▂····▃·▃▅▃█ beats 9/14very easy · avg 86%▇ 0.9%blunt
RVCBench LibriTTS clean voice cloning · Whisper-medium WER7.0% ✎▂▂▂··▂·▃▃▃█▂ beats 7/15avg 8.6%▇ 0.2%blunt · noisy

$ where to run it

Other ways to run it

  1. Run it as published · needs ~0.79 GB · from github.com/myshell-ai
    conda create -n openvoice python=3.9 && conda activate openvoice && git clone https://github.com/myshell-ai/OpenVoice.git && cd OpenVoice && pip install -e . && pip install git+https://github.com/myshell-ai/MeloTTS.git && python -m unidic download && huggingface-cli download myshell-ai/OpenVoiceV2 --revision f36e7edfe1684461a8343844af60babc2efbb727 --local-dir checkpoints_v2 && python -c "import os, torch from openvoice import se_extractor from openvoice.api import ToneColorConverter ckpt_converter = 'checkpoints_v2/converter' device = 'cuda:0' if torch.cuda.is_available() else 'cpu' output_dir = 'outputs_v2' tone_color_converter = ToneColorConverter(f'{ckpt_converter}/config.json', device=device) tone_color_converter.load_ckpt(f'{ckpt_converter}/checkpoint.pth') os.makedirs(output_dir, exist_ok=True) reference_speaker = 'resources/example_reference.mp3' target_se, audio_name = se_extractor.get_se(reference_speaker, tone_color_converter, vad=True) from melo.api import TTS language, text = 'EN_NEWEST', 'Did you ever hear a folk tale about a giant turtle?' src_path = f'{output_dir}/tmp.wav' speed = 1.0 model = TTS(language=language, device=device) speaker_ids = model.hps.data.spk2id for speaker_key in speaker_ids.keys(): speaker_id = speaker_ids[speaker_key] speaker_key = speaker_key.lower().replace('_', '-') source_se = torch.load(f'checkpoints_v2/base_speakers/ses/{speaker_key}.pth', map_location=device) if torch.backends.mps.is_available() and device == 'cpu': torch.backends.mps.is_available = lambda: False model.tts_to_file(text, speaker_id, src_path, speed=speed) save_path = f'{output_dir}/output_v2_{speaker_key}.wav' tone_color_converter.convert(audio_src_path=src_path, src_se=source_se, tgt_se=target_se, output_path=save_path, message='@MyShell') "

Weights and documentation: huggingface.co/myshell-ai ↗

$ how the score was worked out · TTSDS2 listening test results… 73%, TTSDS2 multilingual benchmark… 13%, RVCBench: Benchmarking Robust… 10% and 1 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▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 73%MIT
TTSDS2 multilingual benchmark results (ttsds/leaderboard_data)2▇▇▇▇ 13%MIT
RVCBench: Benchmarking Robustness of Voice Cloning Across Modern Audio Generation Models (arXiv:2602.00443v2)2▇▇▇ 10%CC BY 4.0
TTSDS v1 published results1▇ 1%MIT
prior · pull toward the middle▇ 4%
weighted average= 25.7

$ 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