Talk Converter

A powerful CLI tool that helps you split long YouTube videos into individual talk segments, complete with transcripts and AI-generated descriptions.

Setup

1. Prerequisites

   • Node.js (v16 or higher)
   • ffmpeg (required for video processing)
   • AWS Account (for transcription services)
   • Anthropic API Key (for AI-powered summaries)

2. Installation

   # Clone the repository
   git clone [repository-url]
   cd talk-converter
   
   # Install dependencies
   npm install

3. Environment Variables Copy the .env.template file to .env and fill in the values:

   # Anthropic API Key for generating summaries and articles
   ANTHROPIC_API_KEY=your_anthropic_api_key
   
   # AWS Credentials
   AWS_ACCESS_KEY_ID=your_aws_access_key_id
   AWS_SECRET_ACCESS_KEY=your_aws_secret_access_key
   AWS_REGION=us-east-1              # The AWS region for your services
   AWS_S3_BUCKET=your_s3_bucket_name # The S3 bucket for storing audio files and transcripts
   

   Make sure your AWS credentials are set up correctly with the following permissions:

   • AmazonS3FullAccess
   • AmazonTranscribeFullAccess

   The S3 bucket should be created in your AWS account and accessible with your credentials.

How to Use

1. Basic Usage

   Make sure you create 2 folders in the proejct directory. __talks for talk files __youtube for youtube source videos .mp4

   Use either Youtube Creator studio or the youtube-dl CLI tool download the .mp4 720p minumum video you desire (DEVx stream in this case) and renmae the download to be the {YouTube Video ID}.mp4 ie: youtube.com/watch?v=2cMzN_4guQ0 becomes 2cMzN_4guQ0.mp4

   Then watch the video through and update the run.sh bash script with a single line for each talk you'd like to process. Script is run once per individual talk to be sliced.

   • YouTube URL: Full URL of the YouTube video
   • timestamps: Format "start,end" in seconds or HH:MM:SS format
   • title: Title for the extracted talk segment

2. Example

   npm start -- "https://youtube.com/watch?v=example" "00:15:30,01:45:20" "Understanding AI Systems"

3. Output The tool will create:

   • A video file of the extracted segment (__talks/[title].mp4)
   • An audio transcript (__talks/[title].txt)
   • A markdown file with AI-generated description and article (__talks/[title].md)

4. Processing Steps

   • Downloads the full YouTube video
   • Extracts the specified segment
   • Generates transcript using AWS Transcribe
   • Creates AI-powered summary and article using Claude

Notes

• The tool caches downloaded videos and generated transcripts to avoid reprocessing
• Make sure you have sufficient AWS permissions for S3 and Transcribe services
• Video segments are saved in the __talks directory
• Original downloaded videos are stored in the __youtube directory
• The tool originally was supposed to download the YouTube video for you via code, but I was unable to make that work with the given time constraints