This project provides an end-to-end pipeline for summarizing long YouTube videos using AI, targeting researchers and power users who need to quickly extract knowledge from extensive video content. It automates the process of downloading, transcribing, diarizing, and summarizing video audio, delivering concise, digestible summaries.

How It Works

The system leverages a modular approach, starting with yt-dlp for audio extraction and ffmpeg for decompression. Speech-to-text is handled by faster-whisper, and speaker diarization is performed using pyannote. A chunker.py script segments the transcribed text for efficient processing by Large Language Models (LLMs), with roller-*.py scripts implementing rolling summarization techniques. can-ai-code facilitates LLM inference, and compare.py prepares outputs for a web-based summary viewer (compare-app.py).

Quick Start & Requirements

• Install via pip.
• Requires Python 3.x, yt-dlp, ffmpeg, faster-whisper, and pyannote.
• Setup time and resource requirements are not specified but likely involve significant processing for transcription and LLM inference.

Highlighted Details

• End-to-end pipeline for video summarization.
• Utilizes faster-whisper for efficient speech-to-text.
• Employs pyannote for speaker diarization.
• Features rolling summarization for long-form content.

Maintenance & Community

The project is under active development. No specific community channels or contributor details are provided in the README.

Licensing & Compatibility

The license is not specified in the README. Compatibility for commercial use or closed-source linking is not detailed.

Limitations & Caveats

This project is explicitly stated to be under active development and not ready for production use. Specific limitations or known issues are not detailed.