video_summarization

YouTube Video Summarizer with Indexify

This project demonstrates how to build a YouTube video summarization pipeline using Indexify. The pipeline downloads a YouTube video, extracts audio, transcribes it, classifies the content, and generates a summary based on the classification.

Features

• YouTube video download
• Audio extraction from video
• Speech-to-text transcription using Faster Whisper
• Content classification using Llama.cpp (e.g., job interview, sales call)
• Generate summaries based on conversation type

Prerequisites

• Python 3.9+
• Docker and Docker Compose (for containerized setup)

Installation and Usage

Option 1: Local Installation - In Process

1. Clone this repository:

   git clone https://github.com/tensorlakeai/indexify
   cd indexify/examples/video_summarization
   

2. Create a virtual environment and activate it:

   python -m venv venv
   source venv/bin/activate
   

3. Install the required dependencies:

   pip install -r requirements.txt
   

4. Run the main script:

   python workflow.py --mode in-process-run
   

Option 2: Using Docker Compose - Deployed Graph

1. Clone this repository:

   git clone https://github.com/tensorlakeai/indexify
   cd indexify/examples/video_summarization
   

2. Ensure Docker and Docker Compose are installed on your system.

3. Build the Docker images:

   indexify-cli build-image workflow.py download_youtube_video
   indexify-cli build-image workflow.py extract_audio_from_video
   indexify-cli build-image workflow.py transcribe_audio
   indexify-cli build-image workflow.py classify_meeting_intent
   indexify-cli build-image workflow.py summarize_job_interview
   indexify-cli build-image workflow.py summarize_sales_call
   

4. Start the services:

   docker-compose up --build
   

5. Deploy the graph:

   python workflow.py --mode remote-deploy
   

6. Run the workflow:

   python workflow.py --mode remote-run
   

How it Works

1. Video Processing:

   • Video Download: Uses pytubefix to download the YouTube video.
   • Audio Extraction: Extracts audio from the video using pydub.
   • Transcription: Converts speech to text using Faster Whisper.

2. Content Analysis:

   • Classification: Uses Llama.cpp to classify the content of the transcription.
   • Summarization: Generates a summary based on the classification (job interview or sales call) using Llama.cpp.

Indexify Graph Structure

The project uses the following Indexify graph:

download_youtube_video -> extract_audio_from_video -> transcribe_audio -> classify_meeting_intent -> route_transcription_to_summarizer -> summarize_job_interview
                                                                                                                                       -> summarize_sales_call

Customization

• Modify the youtube_url in the run_workflow() function to process different videos.
• Adjust the classification logic in classify_meeting_intent() to handle more content types.
• Fine-tune the prompts in the summarization functions for better results.
• Experiment with different Llama model variants for improved performance.