🎙️ VibeVoice Studio

A beautiful, modern web application for AI-powered voice synthesis using Microsoft's VibeVoice model. Generate natural-sounding speech from text with custom voice profiles.

✨ Features

• 🎤 Voice Training: Upload audio files or record your voice directly
• 📝 Text-to-Speech: Convert text or text files to natural speech
• 🎭 Multiple Speakers: Support for up to 4 distinct speakers
• 💾 Voice Library: Save and manage custom voice profiles
• 🎨 Beautiful UI: Modern, responsive design with dark/light themes
• ⚡ Real-time Processing: Fast speech generation with streaming support
• 📊 Audio Visualization: Live waveform display during recording
• 💾 Download & Save: Export generated audio files

🎬 Demo

vibevoice-demo.mp4

VibeVoice Studio end-to-end TTS, voice library, and multi-speaker demo.

🚀 Quick Start

Prerequisites

• Python 3.9 or higher
• CUDA-capable GPU (recommended)
• 8GB+ RAM

Installation

1. Clone the repository

git clone https://github.com/shamspias/vibevoice-studio.git
cd vibevoice-studio

2. Create virtual environment

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install VibeVoice

git clone https://github.com/shamspias/VibeVoice
cd VibeVoice
pip install -e .
cd ..

4. Install dependencies

pip install -r requirements.txt

5. Configure environment

cp .env.example .env
# Edit .env with your settings

6. Run the application

python -m app.main

7. Open in browser

http://localhost:8000

🎨 Features Overview

Voice Management

• Upload or record voices
• Support for WAV, MP3, M4A, FLAC
• Organized voice library

Text Processing

• Manual input or upload .txt files
• Multi-speaker support for conversations

Generation Settings

• Voice strength (CFG scale 1.0–2.0)
• Up to 4 speakers
• Adjustable inference steps

Output Options

• Play in browser
• Download WAV file
• Save to library

🔧 Configuration

Edit .env:

HOST=0.0.0.0
PORT=8000
DEBUG=False
MODEL_PATH=microsoft/VibeVoice-1.5B
DEVICE=cuda
CFG_SCALE=1.3
SAMPLE_RATE=24000

🎯 Usage Examples

Basic TTS

1. Select/upload a voice
2. Enter text
3. Click "Generate Speech"

Multi-Speaker

Speaker 1: Hello, welcome!
Speaker 2: Thanks, glad to be here.

Voice Cloning

1. Record 10–30s of clear speech
2. Save with name
3. Use for TTS generation

🛠️ API Documentation

Endpoints

• GET /api/voices — list voices
• POST /api/voices/upload — upload voice
• POST /api/voices/record — record voice
• POST /api/generate — generate speech
• GET /api/audio/{filename} — download audio

🚦 System Requirements

Minimum: Python 3.9+, 8GB RAM, CPU with AVX Recommended: Python 3.10+, 16GB RAM, NVIDIA GPU (8GB+ VRAM)

🐛 Troubleshooting

• OOM: Use smaller model, reduce batch size
• Low quality: Use better voice samples, adjust CFG scale
• Slow generation: Enable GPU, shorten text

📈 Performance Tips

• Use GPU for 10–20× speed
• Batch texts
• Cache voices
• Try quantized models

🤝 Contributing

1. Fork repo
2. Create feature branch
3. Commit & push
4. Open PR

📄 License

MIT License

🙏 Acknowledgments

• Microsoft VibeVoice team
• FastAPI community
• Contributors & users

📞 Support

• Issues: GitHub Issues

🔗 Links

• VibeVoice Model
• FastAPI Docs