🚀 VibeVoice: A Frontier Long Conversational Text-to-Speech Engine, Portable Version

🔧 About

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 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.

✅ Models

Model	Context Length	Generation Length	Weight
VibeVoice-1.5B	64K	~90 min	HF link
VibeVoice-Large	32K	~45 min	HF link
VibeVoice-Q8	32K	~45 min	HF link

⚙️ Installation

VibeVoice uses Python 3.11 and Torch 2.5.1 Cuda 12.1 or 2.7.1 Cuda 12.8.

VibeVoice supports GTX and RTX cards, including GTX10xx/16xx and RTX 20xx–50xx.

🖥️ Windows Portable Installation

This project provided with only *.bat installer/re-installer/starter/updater file, that will download and install all components and build fully portable VibeVoice.

➤ Please Note: - I'm supporting only nVidia GTX10xx/16xx and RTX20xx-50xx GPUs. - This installer is intended for those running Windows 10 or higher. - Application functionality for systems running Windows 7 or lower is not guaranteed.

• Download the VibeVoice .bat installer for Windows in Releases.
• Place the BAT-file in any folder in the root of any partition with a short Latin name without spaces or special characters and run it.
• Select INSTALL (3) entry. Choose desired Torch version to fit your GPU model. .bat file will download, unpack and configure all needed environment.
• After installing, select START (1) or START (2) depending mode you need. .bat will launch Browser, and loads necessary files, models (at forst time).
• UPDATE (4) option updates project from this Git for new version (if it will be).

⚙️ New Features:

• added VibeVoice Q8 model. Works with supported GPUs (RTX20xx-50xx)
• added Ruacent stress marks setup/processing button for russian language
• added additional voices download option (choosable)
• added pre-uploaded reference voice pack (838 voices)
• added INT4 (4Bit) on-the-fly quantize mode to run VibeVoice Large model on PCs with LowVRAM and 32GB+ RAM
• added GPU auto-detection and acceleration auto-select (SDPA/FlashAttention2)
• added autoplay selection
• added random seed generations

📺 Credits

• LeeAeron — additional code, modding, reworking, repository, Hugginface space, features, installer/launcher, reference audios.
• NerualDreming — projecting, first installler version
• Slait - Q8 base code support, help with additional code

📝 License

The VibeVoice code is released under MIT License.