README
VibeVoice Multi-TTS App
Multi-speaker Text-to-Speech application using Microsoft’s VibeVoice models. Supports different model sizes (1.5B, 7B Large, and Large variants).
Prerequisites
-
Python 3.9+
-
FFmpeg must be installed and available in PATH
-
Verify with:
ffmpeg -version
Installation
Clone the repository:
git clone https://github.com/YASSERRMD/vibevoice-multi-tts.git
cd vibevoice-multi-tts
Install dependencies:
pip install -r requirements.txt
Running the App
Start the application with:
python app.py
VRAM Requirements
Different models have different GPU memory (VRAM) needs. Choose the model based on your available GPU.
| Model Variant | Typical VRAM Usage | Recommended VRAM | Notes |
|---|---|---|---|
| VibeVoice 1.5B | ~7 GB | 8 GB | Handles up to 90-min speech with 64K context. |
| VibeVoice 7B | ~17–19 GB | 24 GB | Higher fidelity, ~45-min speech with 32K context. |
| VibeVoice Large | ~19–20 GB | 28 GB | Full ~9B model, optimized for long-form synthesis. |
Model Selection
- Default: 1.5B (lightweight, widely compatible).
- For higher-quality output: use 7B Large.
- For mid-range GPUs: use Large-Quant-4bit.
Notes
- Ensure FFmpeg is installed before running, otherwise audio processing will fail.
- Quantized models significantly reduce VRAM usage and storage size.
- For multi-hour audio generation, ensure sufficient disk space and GPU memory.
Would you like me to also add example commands/code snippets (like how to switch between 1.5B and 7B inside app.py) so users can quickly try different models?
Contributors
11 commits
Similar repos
Languages
Python
100.0%
README
VibeVoice Multi-TTS App
Multi-speaker Text-to-Speech application using Microsoft’s VibeVoice models. Supports different model sizes (1.5B, 7B Large, and Large variants).
Prerequisites
-
Python 3.9+
-
FFmpeg must be installed and available in PATH
-
Verify with:
ffmpeg -version
Installation
Clone the repository:
git clone https://github.com/YASSERRMD/vibevoice-multi-tts.git
cd vibevoice-multi-tts
Install dependencies:
pip install -r requirements.txt
Running the App
Start the application with:
python app.py
VRAM Requirements
Different models have different GPU memory (VRAM) needs. Choose the model based on your available GPU.
| Model Variant | Typical VRAM Usage | Recommended VRAM | Notes |
|---|---|---|---|
| VibeVoice 1.5B | ~7 GB | 8 GB | Handles up to 90-min speech with 64K context. |
| VibeVoice 7B | ~17–19 GB | 24 GB | Higher fidelity, ~45-min speech with 32K context. |
| VibeVoice Large | ~19–20 GB | 28 GB | Full ~9B model, optimized for long-form synthesis. |
Model Selection
- Default: 1.5B (lightweight, widely compatible).
- For higher-quality output: use 7B Large.
- For mid-range GPUs: use Large-Quant-4bit.
Notes
- Ensure FFmpeg is installed before running, otherwise audio processing will fail.
- Quantized models significantly reduce VRAM usage and storage size.
- For multi-hour audio generation, ensure sufficient disk space and GPU memory.
Would you like me to also add example commands/code snippets (like how to switch between 1.5B and 7B inside app.py) so users can quickly try different models?
Similar repos
Contributors
11 commits
Languages
Python
100.0%