whatmodel.sh / tts / VibeVoice 1.5B

VibeVoice 1.5B

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

VibeVoice 1.5B is an open-weights text-to-speech model from Microsoft, released August 2025. Its published size is 2.7B parameters. Licence: mit.

53.3±6.0#57 of 118 text-to-speech modelsfrom 3 benchmark results across 2 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 VibeVoice 1.5B, out of 148 with a score, then more from Microsoft.

#modelscorewhereRAMreleased
68CosyVoice-300M Alibaba (FunAudioLLM)54±3local~2.3GB2024-07
69Chatterbox Multilingual V3 Resemble AI54±10local~3.2GB2026-04
70VibeVoice 1.5B Microsoft53±6local~5.4GB2025-08
71Lightning v3.1 Smallest.ai53±3api $17.50/M chars2026-03
72Gepard 1.0 nineninesix.ai53±4comparison only2026-06
more from Microsoft
81Azure Ameha Neural Microsoft50±16comparison only2021-10
82Azure Mekdes Neural Microsoft50±16comparison only2021-10
140SpeechT5 TTS Microsoft23±3local~0.7GB2023-02

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

$ what it's good at

other tests3 not in a task · 85% of its score
testresultvs the others in its rundifficultycountswhy
RVCBench LibriTTS clean voice cloning · ECAPA speaker similarity48.0% ✎▃·····▂▂▃▃█▆ beats 4/15medium · avg 46%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 52%blunt · at its level
VibeVoice podcast · preference MOS (MIT figure) lab3.43 / 5 ✎█·█········█ beats 1/2medium · avg 3.08 / 5▇▇▇▇▇▇▇▇▇▇▇▇▇ 33%at its level · small run
RVCBench LibriTTS clean voice cloning · Whisper-medium WER23.0% ✎▂▂▃··▂·▃▃▃█▂ beats 2/15avg 8.6%▇ 0.6%blunt · noisy

$ where to run it

Other ways to run it

  1. Run it as published · needs ~5.42 GB · from github.com/huggingface
    pip install "transformers[torch]>=5.17.0" diffusers soundfile && python -c "import os; from transformers import AutoProcessor, AutoModelForTextToWaveform; model_id='vibevoice/VibeVoice-1.5B-hf'; processor=AutoProcessor.from_pretrained(model_id); model=AutoModelForTextToWaveform.from_pretrained(model_id, device_map='auto'); conversation=[{'role':'0','content':[{'type':'text','text':'Hello, nice to meet you. How are you?'}]}]; inputs=processor.apply_chat_template(conversation, return_dict=True, tokenize=True, add_generation_prompt=True).to(model.device, model.dtype); audio=model.generate(**inputs); processor.save_audio(audio, 'VibeVoice-1.5B-hf_tts.wav')"

Official weights and documentation: huggingface.co/microsoft ↗

$ how the score was worked out · RVCBench: Benchmarking Robust… 52%, Microsoft VibeVoice own subje… 33%

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
RVCBench: Benchmarking Robustness of Voice Cloning Across Modern Audio Generation Models (arXiv:2602.00443v2)2▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 52%CC BY 4.0
Microsoft VibeVoice own subjective evaluation figure · lab-reported1▇▇▇▇▇▇▇▇▇▇ 33%MIT
prior · pull toward the middle▇▇▇▇ 15%
weighted average= 53.3

$ 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