🎤 RVC Training WebUI (VCTrain)

Retrieval-based Voice Conversion - A comprehensive training system with web interface for custom voice model training.

This project is based on PolTrain by Politrees
Side project of RVC Starter

---

🚀 Quick Start

Launch the WebUI

# Install dependencies
pip install gradio torch torchaudio

# Launch WebUI
python webui/launch.py

Then open your browser to http://127.0.0.1:7860

---

📋 Features

🎯 11+ Optimizers

Choose from a wide range of optimization algorithms:

• AdamW - Default, stable
• Adam - Classic adaptive
• AdaBelief - Fast convergence
• AdaBeliefV2 - With AMSGrad
• Adafactor - Memory efficient
• AMSGrad - Prevents oscillations
• SGD - With Nesterov momentum
• RAdam - Rectified Adam
• Lion - Sign-based, memory efficient
• AdamP - Better generalization
• Sophia - Second-order clipping

🌐 Gradio WebUI

Complete web interface with 6 tabs:

Tab	Description
🏠 Home	Dashboard, system info, quick start guide
📊 Data Preprocessing	Audio processing, feature extraction
⚙️ Training Config	Model architecture, hyperparameters
🚀 Model Training	Real-time training, monitoring
🎵 Voice Conversion	Inference, pitch adjustment
📁 Model Management	Export, delete, organize models

---

📁 Project Structure

VCTrain/
├── webui/                      # Gradio WebUI
│   ├── app.py                  # Main application
│   ├── launch.py               # Launcher script
│   ├── requirements.txt        # Dependencies
│   ├── README.md               # WebUI documentation
│   └── tabs/                   # Tab modules
│       ├── home_tab.py
│       ├── data_preprocessing_tab.py
│       ├── training_config_tab.py
│       ├── model_training_tab.py
│       ├── inference_tab.py
│       └── model_management_tab.py
│
├── rvc/
│   ├── train/
│   │   ├── train.py            # Training script
│   │   ├── utils/
│   │   │   └── optimizers/     # 11 optimizer implementations
│   │   ├── preprocess/         # Data preprocessing
│   │   └── ...
│   ├── lib/                    # Core libraries
│   └── configs/                # Configuration files
│
├── experiments/                # Training outputs (created automatically)
└── logs/                       # Training logs

---

🔧 Usage

Command Line Training

python rvc/train/train.py \
  --experiment_dir "experiments" \
  --model_name "my_voice" \
  --optimizer "AdamW" \
  --total_epoch 300 \
  --batch_size 8 \
  --sample_rate 48000 \
  --gpus "0"

WebUI Launch Options

# Default
python webui/launch.py

# Custom port
python webui/launch.py --port 7861

# Public share link
python webui/launch.py --share

# With authentication
python webui/launch.py --auth username:password

---

⚙️ Optimizer Guide

Recommended Settings

Optimizer	Learning Rate	Best For
AdamW	1e-4	Default choice
AdaBelief	1e-4	Fast convergence
Adafactor	Auto	Low VRAM
Lion	1e-5	Memory efficient
Sophia	5e-5	Stable training

Performance Comparison

Optimizer	Speed	Quality	VRAM Usage
AdamW	⭐⭐⭐	⭐⭐⭐	⭐⭐
AdaBelief	⭐⭐⭐	⭐⭐⭐⭐	⭐⭐
Adafactor	⭐⭐	⭐⭐⭐	⭐⭐⭐⭐
Lion	⭐⭐⭐⭐	⭐⭐⭐	⭐⭐⭐
Sophia	⭐⭐⭐⭐	⭐⭐⭐⭐	⭐⭐

---

📊 Training Workflow

1. Prepare Data → Collect clean audio files (WAV, 32kHz+)
2. Preprocess → Use Data Preprocessing tab
3. Configure → Set parameters in Training Config tab
4. Train → Start training in Model Training tab
5. Convert → Use trained model in Voice Conversion tab

---

💡 Tips

Dataset Quality

• Use clean audio without background noise
• Minimum 10 minutes of speech recommended
• Consistent volume levels
• Remove silence and breaths

Training

• Start with 100 epochs for testing
• Use 300+ epochs for production
• Monitor loss values (should decrease)
• Target mel similarity: 70%+

VRAM Optimization

• Use Adafactor for low VRAM
• Reduce batch size
• Enable gradient checkpointing
• Use FP16 mixed precision

---

🛠️ Requirements

• Python: 3.8+
• PyTorch: 2.0+
• GPU: CUDA 11.7+ (optional, CPU supported)
• RAM: 8GB+ recommended

Core Dependencies

torch>=2.0.0
torchaudio>=2.0.0
gradio>=4.0.0
librosa>=0.10.0
tensorboard>=2.13.0

---

📚 Documentation

• WebUI README - Detailed WebUI documentation
• Optimizers README - Optimizer guide

---

🙏 Acknowledgments

• PolTrain - Base project
• RVC - Voice conversion technology
• Gradio - Web UI framework
• PyTorch - Deep learning framework

---

📝 License

Same license as the original PolTrain project.

---

Happy Training! 🎤