Audiobook Generator

A high-quality text-to-speech converter that transforms PDF and TXT files into MP4 audio files. Currently using WhisperSpeech technology, but adaptable to future better models.

Example Output

Example.mp4

Features

• Supports PDF and TXT input files
• GPU acceleration with CUDA support
• High-quality audio output (44.1kHz, 320kbps AAC)
• Efficient memory management and batch processing
• Multi-threaded CPU processing

Requirements

• Python 3.x
• NVIDIA GPU with CUDA support (optional)
• Minimum 4GB RAM
• Required packages listed in requirements.txt

Installation

1. Clone the repository
2. Install dependencies:

   pip install -r requirements.txt

Usage

1. Place PDF files in PDF directory
2. Place TXT files in TXT directory
3. Run:

   python main.py

4. Find generated audio in Audio directory

Technical Details

Performance Optimizations

• CUDA-aware processing with automatic GPU detection
• Dynamic batch sizing based on available VRAM/RAM
• Multi-threaded CPU processing for non-GPU operations
• Memory-efficient chunking for large documents

Audio Processing

• 44.1kHz sampling rate
• 320kbps AAC encoding
• Stereo output
• Zero-quality loss audio settings

System Architecture

• Modular pipeline design for easy model swapping
• Buffered I/O operations (1MB buffer)
• Automatic memory management with CUDA cache clearing
• Fault-tolerant processing with error handling

Resource Management

• Dynamic worker allocation based on system specs
• Configurable chunk sizes (default: 2000 tokens)
• Adaptive batch processing
• Progressive audio concatenation

Contributing

Feel free to suggest optimizations or improvements through issues or pull requests. The system is designed to be modular, allowing for easy integration of new TTS models.