PDF to Speech (PTS)

This project converts PDF documents into speech audio files using a Text-to-Speech (TTS) model. It extracts text from a PDF, splits it into manageable chunks, and synthesizes each chunk into an audio file using the Coqui TTS library.

Features

• Extracts and cleans text from PDF files (page by page)
• Supports chunking of text for long documents
• Synthesizes audio using customizable TTS models (including language and speaker options)
• Outputs audio files (WAV) for each chunk/page

Requirements

• Python 3.8+
• Coqui TTS
• pdfminer.six
• torch

Installation

1. Clone this repository:

   git clone https://github.com/mtb0x1/pdf_to_speech
   cd pdf_to_speech

2. Install dependencies:

   pip install TTS==0.13.0 pdfminer.six==20221105 torch==2.0.1

   You may need to install additional dependencies for your TTS model (see Coqui TTS documentation). For GPU acceleration, install the CUDA-enabled version of torch.

Usage

Run the script from the command line:

python process.py --pdf <path_to_pdf> [--model <tts_model>] [--num-pages <N>] [--language <lang>] [--speaker <name>] [--speaker-wav <wav_path>] [--log-level <level>]

Arguments

• --pdf (required): Path to the PDF file to convert.
• --model: TTS model to use (default: tts_models/fr/css10/vits).
• --num-pages: Number of pages to process (default: all pages).
• --language: Language code for TTS synthesis (e.g., fr-fr).
• --speaker: Speaker name (if supported by the model).
• --speaker-wav: Path to a reference speaker WAV file (if supported).
• --log-level: Logging level (debug, info, warning, error).

Example

Convert the first 5 pages of a French PDF to audio using the tts_models/fr/css10/vits model:

python process.py --pdf file.pdf --model tts_models/fr/css10/vits --num-pages 5

The resulting audio files will be saved in the output/ directory as page_001.wav, page_002.wav, etc.

Notes

• For best results, use a TTS model that matches the language of your PDF.
• Some models support multiple speakers or require additional arguments (see Coqui TTS documentation).
• If you encounter CUDA errors, ensure your system has a compatible GPU and CUDA drivers, or the script will fall back to CPU.

License

MIT License