transcribeexx_skill

Overview

This skill transcribes YouTube videos and local audio/video files with speaker diarization and produces clean, speaker-labeled transcripts ready for LLM analysis. It relies on OpenClawCLI and integrates ElevenLabs for diarization quality. Outputs include plain and raw transcripts, word-level timings, and metadata organized into timestamped folders.

How this skill works

The tool downloads or reads the media file, runs speech-to-text with speaker diarization, and saves multiple output artifacts (labeled transcript, raw text, JSON timings, metadata). It uses yt-dlp/FFmpeg for media handling and OpenClawCLI as the front-end orchestrator; ElevenLabs is used to improve speaker separation. Final files are stored under ~/Documents/transcripts/{category}/{title}-{date}/ for easy retrieval and LLM input.

When to use it

• Transcribing a YouTube URL for research or summarization
• Converting podcast episodes into searchable, speaker-labeled text
• Preparing interview or meeting recordings for analysis
• Generating word-level timings for alignment or subtitle creation
• Archiving audio/video with metadata for later LLM processing

Best practices

• Install OpenClawCLI before use and confirm API keys for ElevenLabs in the tool's .env
• Quote URLs that contain & or special characters when passing them to the command
• Ensure yt-dlp and ffmpeg are installed and in PATH (e.g., brew install yt-dlp ffmpeg on macOS)
• Validate audio quality and channel configuration for best diarization (clean audio and minimal overlap)
• Use the generated metadata.json to route transcripts into downstream LLM pipelines

Example use cases

• Transcribe a YouTube lecture to generate study notes and timestamps
• Convert a podcast episode into a speaker-labeled transcript for editing and SEO
• Process recorded interviews to separate questions and answers for analysis
• Create subtitle-ready word-level timing files for video localization
• Archive batches of videos with metadata for a searchable media library