YouTube Audio Transcription and Summarization Tool

A Python-based tool to download audio from YouTube videos, transcribe the audio using Faster Whisper, and generate concise summaries with a locally hosted LLaMA language model.

Table of Contents

• Features
• Prerequisites
• Installation
• Setting Up llama.cpp Server
• Downloading the Language Model
• Usage
• Example
• Troubleshooting
• Contributing
• License

Features

• Download Audio: Extracts audio from YouTube videos in MP3 format.
• Transcription: Utilizes Faster Whisper with GPU acceleration for efficient and accurate transcription.
• Summarization: Generates concise summaries using a locally hosted LLaMA language model.
• Token Counting: Provides the number of tokens in the transcription for API usage management.
• User-Friendly Output: Displays summaries in Markdown format using the rich library for enhanced readability.

Prerequisites

Before using this tool, ensure you have the following installed on your system:

• Python 3.8+: Ensure you have Python installed. You can download it from python.org.

• FFmpeg: Required for audio processing.

  • Ubuntu/Debian:

  sudo apt update
  sudo apt install ffmpeg

  • macOS (using Homebrew):

  brew install ffmpeg

  • Windows: Download the latest FFmpeg build from FFmpeg Downloads. Follow the installation instructions for your system.

• CUDA: If you have an NVIDIA GPU and wish to utilize GPU acceleration for Faster Whisper, ensure CUDA is installed and properly configured. Refer to the CUDA Installation Guide for details.

• Git: To clone repositories.

  • Ubuntu/Debian:

  sudo apt install git

  • macOS (using Homebrew):

  brew install git

  • Windows: Download and install Git from git-scm.com.

Installation

Clone the Repository

git clone https://github.com/nmandic78/yt_summary.git
cd yt_summary

Create a Virtual Environment (Optional but Recommended)

python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

Install Python Dependencies

Ensure you have pip updated:

pip install --upgrade pip

Install the required packages:

pip install -r requirements.txt

or:

pip install yt-dlp faster-whisper openai tiktoken rich

Setting Up llama.cpp Server

To generate summaries, the tool relies on a locally hosted LLaMA language model using llama.cpp. Follow the steps below to set up the server.

Clone the llama.cpp Repository

git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp

Build the Server

Ensure you have the necessary build tools installed (e.g., make, gcc).

make

This will compile the llama-server executable.

Download a LLaMA Model from Hugging Face

• Visit the Hugging Face Models page.
• Search for a compatible LLaMA model, such as gemma-2-9b-it-Q8_0.gguf.
• Download the model and place it in a directory of your choice, e.g., /mnt/disk2/LLM_MODELS/models/gemma-2-9b-it-Q8_0.gguf.

Note: Ensure you have the rights and necessary permissions to use the model.

Run the llama.cpp Server

Execute the server with your chosen model:

./llama-server -m /mnt/disk2/LLM_MODELS/models/gemma-2-9b-it-Q8_0.gguf -ngl 99 -c 8192

Parameters Explained:

• -m: Path to the model file.
• -ngl: Number of GPU layers (adjust based on your GPU capabilities).
• -c: Context size in tokens (adjust as needed).

The server will start and listen on http://localhost:8080/v1.

Downloading the Language Model

If you haven't downloaded a LLaMA model yet, follow the steps in the Setting Up llama.cpp Server section to obtain a compatible model from Hugging Face.

Usage

Once you have set up the llama.cpp server and installed all dependencies, you can use the transcription and summarization tool.

Command-Line Arguments

• -v, --video_url: (Required) YouTube video URL to download and transcribe.
• -m, --mp3_dir: (Optional) Directory to save the downloaded MP3 file. Default: /home/yourusername/Music/YT_AUDIOS/
• -t, --transcript_dir: (Optional) Directory to save the transcription text file. Default: /home/yourusername/Music/YT_AUDIOS/

Running the Script

python yt_summary.py -v <YouTube_Video_URL> [options]

Example:

python yt_summary.py -v https://www.youtube.com/watch?v=dQw4w9WgXcQ

This command will:

• Download the audio from the provided YouTube video URL and save it as an MP3 file in the default directory.
• Transcribe the audio using Faster Whisper.
• Generate a summary using the locally hosted LLaMA model.
• Display the summary in the console and save the transcription to a text file.

Specifying Custom Directories

You can specify custom directories for saving MP3 files and transcriptions:

python yt_summary.py -v <YouTube_Video_URL> -m /path/to/mp3_dir -t /path/to/transcript_dir

Example summary

Troubleshooting

• FFmpeg Not Found: Ensure FFmpeg is installed and added to your system's PATH.
• CUDA Issues: Verify that CUDA is correctly installed and that your GPU supports the required operations.
• llama.cpp Server Not Running: Ensure the server is running before executing the transcription script. Verify the server URL and port.
• Missing Dependencies: Ensure all Python packages are installed. Re-run pip install -r requirements.txt if necessary.
• Insufficient Permissions: Check directory permissions for saving MP3 and transcription files.

Contributing

Contributions are welcome! Please follow these steps:

Fork the Repository

Create a Feature Branch

git checkout -b feature/YourFeature

Commit Your Changes

git commit -m "Add YourFeature"

Push to the Branch

git push origin feature/YourFeature

Open a Pull Request

Please ensure your code follows the project's coding standards and includes appropriate documentation.

License

This project is licensed under the MIT License.

Developed by Nenad Mandic