This tool provides a command-line interface for transcribing audio and video content, primarily from YouTube URLs. It targets users who need to process spoken content for text-based analysis, offering features like transcription, summarization, and translation, leveraging local AI models and optional cloud APIs.

How It Works

The application utilizes the Whisper model for local speech-to-text transcription, allowing users to select model sizes for a trade-off between accuracy and performance. For summarization and translation, it integrates with OpenAI or Claude APIs, requiring user-provided API keys. The core workflow involves fetching audio from a URL, processing it with Whisper, and then optionally piping the output to separate summarize and translate scripts.

Quick Start & Requirements

• Installation: Recommended via Nix for deterministic dependency management. Alternatively, manual installation requires bash, python3.12, ffmpeg, and optionally glow.
• Dependencies: Python dependencies are managed via pip within a cached venv. Whisper models are downloaded locally.
• API Keys: OpenAI or Claude API keys are required for summarization and translation features.
• Setup: Nix installation is recommended; setup without Nix is untested on platforms other than Mac.
• Docs: https://github.com/pmarreck/yt-transcriber

Highlighted Details

• Supports transcription of local audio/video files via ffmpeg.
• Offers selectable Whisper model sizes (base, small, medium, large, large-v2).
• Includes companion scripts for summarization and translation.
• Test suite available via yt-transcriber TEST.

Maintenance & Community

The project appears to be a personal utility with limited public community engagement signals in the README.

Licensing & Compatibility

The README does not explicitly state a license.

Limitations & Caveats

Speaker identification is marked as a work-in-progress (WIP). Support for other video platforms and conversion to a web service are future features. Setup has only been tested on macOS with Nix.