Skip to content

Latest commit

 

History

History
110 lines (88 loc) · 3.19 KB

README.md

File metadata and controls

110 lines (88 loc) · 3.19 KB

YouTube Transcript Search and Clip Extraction Tool

A Python tool that combines AssemblyAI transcription with semantic search to find and extract specific moments from YouTube videos.

Prerequisites

  • Python 3.10 or higher
  • AssemblyAI API key (set in .env file)
  • FFmpeg (required for audio/video processing)

Installation

1. Install FFmpeg

macOS

Using Homebrew (recommended):

brew install ffmpeg

For other methods visit FFmpeg website

2. Verify FFmpeg Installation

ffmpeg -version

3. Install Python Dependencies

Using pip:

pip install -r requirements.txt

Using pipenv:

pip install pipenv
pipenv install
pipenv shell

Configuration

Create a .env file with your AssemblyAI API key:

ASSEMBLYAI_AUTH_KEY=your-api-key-here

Usage

Basic Command

python transcript.py <youtube_url> (-p PHRASE | -t TEXT | -s SRT_FILE) [options]

Parameters

Parameter Description Required Default
youtube_url Full URL of the YouTube video Yes -
-p, --phrase Short phrase to search for (max 5 words) No* -
-t, --text Full text to search for (split into segments) No* -
-s, --srt Path to SRT file for subtitle timestamps No* -
--threshold Minimum similarity threshold (0-100) No 80
--clip-duration Duration of extracted clips in seconds (0 to disable) No 30
--no-cleanup Keep temporary files after processing No False
-w, --words Words per subtitle frame (window size) No 1
-f, --font-size Font size for the subtitle text No 72

* One of --phrase, --text, or --srt must be specified, but they are mutually exclusive

Examples

Search for a specific phrase:

python transcript.py "https://youtube.com/watch?v=example" -p "interesting phrase" --clip-duration 30

Search for a longer text segment:

python transcript.py "https://youtube.com/watch?v=example" -t "this is a longer text that will be split into start and end segments" --threshold 90

Search with five-word subtitle window:

python transcript.py "https://youtube.com/watch?v=example" -p "interesting phrase" --clip-duration 30 --words 5

Use a predefined SRT file:

python transcript.py "https://youtube.com/watch?v=example" -s subtitles.srt --words 5

Customize subtitle font size:

python transcript.py "https://youtube.com/watch?v=example" -p "interesting phrase" --font-size 48

Output

The tool will:

  1. Download and transcribe the video audio (or use provided SRT file)
  2. Search for matches using fuzzy string matching (when using -p or -t)
  3. Display matches with:
    • Timestamp
    • Match score (for phrase/text search)
    • Matched text
    • Direct YouTube URL with timestamp
  4. Create video clips with word-by-word subtitles

Notes

  • Search phrases are limited to 5 words for optimal matching
  • Long text inputs are automatically split into start and end segments
  • Temporary files are automatically cleaned up unless --no-cleanup is specified
  • When using an SRT file (-s), phrase searching is disabled and the timestamps from the file are used directly