Skip to content

Latest commit

 

History

History
134 lines (105 loc) · 5.6 KB

README.md

File metadata and controls

134 lines (105 loc) · 5.6 KB

🎙️ PDF to Podcast Conversion with LLMs 📄➡️🎧

This repository hosts a pipeline for transforming PDF files into audio podcasts! Leveraging advanced Language Models (LLMs) and Text-to-Speech (TTS) technology, this tool converts textual PDF information into engaging, high-quality audio content. It’s ideal for creators, researchers, or anyone looking to make static documents more accessible through audio.

📑 Table of Contents

  1. 📋 Project Overview
  2. ✨ Features
  3. 🔧 Installation
  4. 🔄 Workflow
  5. 📝 Detailed Steps
  6. 📋 Requirements
  7. 🚀 Usage
  8. 📂 File Structure
  9. 🙏 Acknowledgments

📋 Project Overview

This project uses modern LLMs and TTS technology to automate the conversion of PDFs into audio podcasts. Built on Llama 3.1 8b and high-quality TTS models (Parler and Bark), this tool delivers a seamless process from raw PDF to professional-sounding podcast with minimal manual input. The modular Jupyter Notebook structure makes each step transparent and easy to customize.

✨ Features

  • 📄 Automated PDF Text Extraction: Quickly extracts text from PDFs, accommodating various formats.
  • 🧠 LLM-Driven Content Generation: Creates scripts that are rephrased and summarized for podcast-style delivery.
  • 📝 Script Refinement: Enhances readability and flow, tailoring content for audio listeners.
  • 🎙️ Natural Sounding TTS Output: Uses Parler and Bark TTS models to produce clear, natural-sounding audio.

🔧 Installation

To set up the environment, ensure Python is installed, then execute:

pip install -r requirements.txt

This installs dependencies, including foundational packages like torch and transformers needed for TTS and NLP tasks.

🔄 Workflow

The project workflow includes four primary stages, each with a dedicated notebook:

  1. Extract Text 📄 from the PDF.
  2. Generate Podcast Script 📝 using Llama 3.1 8b.
  3. Refine Script ✍️ for clarity and flow.
  4. Synthesize Audio 🎧 for podcast-ready MP3 output.

📝 Detailed Steps

📕 Step 1: PDF Text Extraction

  • Notebook: Step1_PDF_preprocessing.ipynb
  • Purpose: Extracts text from PDF documents for easy processing.
  • Libraries: Utilizes PyPDF2 for PDF parsing.
  • Process:
    • Converts each PDF page into cleaned text.
    • Saves extracted text as a .txt file, ready for LLM processing.

✍️ Step 2: Podcast Script Generation

  • Notebook: Step2_Transcript_Writer.ipynb
  • Purpose: Uses Llama 3.1 8b to summarize and rephrase content for audio.
  • Highlights:
    • The LLM converts extracted text into a listener-friendly script.
    • Produces a draft script saved as a text file for refinement.

🛠️ Step 3: Transcript Refinement

  • Notebook: Step3_Rewriter.ipynb
  • Purpose: Polishes the initial script, optimizing it for audio narration.
  • Process:
    • Improves readability and conversational flow.
    • Reduces redundancy and clarifies complex ideas.

🎙️ Step 4: Audio Generation

  • Notebook: Step4_TTS_Workflow.ipynb
  • Purpose: Converts the refined script into high-quality audio.
  • Models Used: Combines Parler and Bark TTS models.
  • Output: Creates an .mp3 file that’s ready for playback or distribution.

📋 Requirements

The requirements.txt file provides a comprehensive list of dependencies, including:

  • NLP and LLM: transformers, torch, datasets
  • PDF Parsing: PyPDF2
  • Audio Processing: pydub, audioread, librosa
  • TTS Models: parler_tts, bark

Key Dependencies

  • LLM Processing: transformers, torch
  • Text Extraction: PyPDF2
  • Audio Tools: pydub

🚀 Usage

  1. Prepare PDF Files 📄: Place PDFs in a folder accessible to the notebooks.
  2. Run Notebooks Sequentially:
    • Step 1: Run Step1_PDF_preprocessing.ipynb to extract text.
    • Step 2: Run Step2_Transcript_Writer.ipynb to create a script.
    • Step 3: Run Step3_Rewriter.ipynb to refine the script.
    • Step 4: Run Step4_TTS_Workflow.ipynb to generate audio.
  3. Review Output 📄: Check the output at each step to ensure quality.

📂 File Structure

The project includes the following files:

  • Step1_PDF_preprocessing.ipynb: PDF text extraction.
  • Step2_Transcript_Writer.ipynb: Script generation.
  • Step3_Rewriter.ipynb: Script refinement.
  • Step4_TTS_Workflow.ipynb: Audio synthesis.
  • requirements.txt: Lists all dependencies.

🙏 Acknowledgments

This project is powered by the amazing open-source tools provided by:

  • Meta AI for the Llama 3.1 8b language model.
  • Parler and Bark TTS for high-quality audio generation.
  • PyPDF2 and Pydub for essential PDF and audio handling.

🔗 General Links & Resources