ayo/VibeVoice
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Update README

Browse source
This commit is contained in:
Zhiliang Peng 2025-08-25 22:50:42 +08:00 committed by GitHub
parent be79602c7f
commit 1dd31995f5
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 19 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

25
README.md
View file

@ -1,19 +1,32 @@
# VibeVoice: A Frontier Open-Source Text-to-Speech Model
<p align="center">
<a href="https://microsoft.github.io/VibeVoice">
<img src="https://img.shields.io/badge/🌐_Project_Page-4285F4?style=for-the-badge&logo=google-chrome&logoColor=white" alt="Project Page">
</a>
<a href="https://huggingface.co/collections/microsoft/vibevoice-68a2ef24a875c44be47b034f">
<img src="https://img.shields.io/badge/🤗_Hugging_Face-FFD21E?style=for-the-badge&logo=huggingface&logoColor=black" alt="Hugging Face">
</a>
<a href="https://aka.ms/VibeVoiceDemo">
<img src="https://img.shields.io/badge/🎵_Demo-FF6B6B?style=for-the-badge&logo=gradio&logoColor=white" alt="Demo">
</a>
</p>
VibeVoice is a novel framework designed for generating **expressive**, **long-form**, **multi-speaker** conversational audio, such as podcasts, from text. It addresses significant challenges in traditional Text-to-Speech (TTS) systems, particularly in scalability, speaker consistency, and natural turn-taking.
A core innovation of VibeVoice is its use of continuous speech tokenizers (Acoustic and Semantic) operating at an ultra-low frame rate of 7.5 Hz. These tokenizers efficiently preserve audio fidelity while significantly boosting computational efficiency for processing long sequences. VibeVoice employs a [next-token diffusion](https://arxiv.org/abs/2412.08635) framework, leveraging a Large Language Model (LLM) to understand textual context and dialogue flow, and a diffusion head to generate high-fidelity acoustic details.
The model can synthesize speech up to **90 minutes** long with up to **4 distinct speakers**, surpassing the typical 1-2 speaker limits of many prior models.
You can try it in our host [Gradio demo](https://aka.ms/VibeVoiceDemo).
Try it out via [Demo](https://aka.ms/VibeVoiceDemo).
## Models
| Model | Base Model | Context Length | Generation Length | Weight |
|-------|------------|----------------|----------|----------|
| VibeVoice-Stream-0.5B | Qwen2.5-0.5B | - | - | On the way |
| VibeVoice-1.5B | Qwen2.5-1.5B | 64K | ~90 min | [HF link](https://huggingface.co/microsoft/VibeVoice-1.5B) |
| VibeVoice-7B | Qwen2.5-7B | 32K | ~45 min | On the way |
| Model | Context Length | Generation Length | Weight |
|-------|----------------|----------|----------|
| VibeVoice-0.5B-Streaming | - | - | On the way |
| VibeVoice-1.5B | 64K | ~90 min | [HF link](https://huggingface.co/microsoft/VibeVoice-1.5B) |
| VibeVoice-7B| 32K | ~45 min | On the way |
## Installation
We recommend to use NVIDIA Deep Learning Container to manage the CUDA environment.