Voice-Clone

🎙️ Voice Cloning Web Application

A production-quality, CPU-optimized Voice Cloning Web Application built for Windows using Resemble AI's Chatterbox TTS and Gradio. The application works completely offline after the initial model download and ensures minimal RAM/CPU footprint.

📁 Project Structure

• app.py: Main Gradio Web UI script.
• install.py: Installer script that automatically sets up the environment, installs dependencies, and downloads the model.
• requirements.txt: List of dependencies.
• download_model.py: Script responsible for downloading the Chatterbox TTS model.
• config.py: Project configuration and paths.
• utils.py: Audio processing (format conversion, silence removal, normalization).
• model_loader.py: Handles loading the TTS model using a memory-optimized Singleton pattern.
• generate.py: Executes the model inference with memory garbage collection.
• outputs/: Saved generated audios.
• uploads/: Processed uploads.
• cache/: Cached Chatterbox model.
• venv/: Virtual environment.

⚙️ Installation

1. Open a Command Prompt or PowerShell terminal.
2. Navigate to this directory.
3. Run the installer:

   python install.py
   

   This will create a virtual environment, install CPU-only PyTorch (to save space), install all dependencies, and pre-download the model.

🚀 Running the App

To make launching as easy as possible, a launcher script is included.

Simply double-click the start.bat file in your folder, or run it from your terminal:

.\start.bat

This script will automatically:

1. Install any missing frontend dependencies.
2. Launch the FastAPI AI Backend in a background window.
3. Start the React Frontend.
4. Open the stunning new Voice Cloning Studio in your browser.

The browser will automatically open your stunning new Voice Cloning Studio at http://localhost:5173.

🛠️ Usage

1. Upload Voice: Upload a WAV/MP3/FLAC file. It must be between 5 and 30 seconds. The app will automatically convert, normalize, and remove silence.
2. Type Text: Enter the text you want the AI to speak.
3. Generate: Click the generate button. Generation time will depend on your CPU capability.
4. Download: The audio will appear on the right side and is automatically saved in the outputs/ folder.

❓ FAQ & Troubleshooting

• Model takes too long to load? The first run might take a while to initialize. Subsequent generations use the cached model in memory.
• Out of Memory Error? Ensure you don't have other heavy applications open. The script automatically uses gc.collect() to free memory.
• Audio Error? Ensure your reference audio is clear, contains only one speaker, and is not overly noisy.