transcribee

Transcribee is an automated transcription and diarization tool designed for content creators, researchers, and developers who need to convert spoken content into structured data for LLM processing. By leveraging yt-dlp and FFmpeg, it extracts audio from YouTube URLs or processes local media files (including mp3, mp4, wav, and mkv) to generate clean, speaker-labeled transcripts. The skill automates the identification of individual speakers using ElevenLabs-powered diarization, ensuring that conversational context is preserved.

• Multi-format support: Process YouTube URLs directly or handle local media files (Audio: mp3, m4a, wav, ogg, flac; Video: mp4, mkv, webm, mov, avi).

• High-fidelity diarization: Differentiates between speakers to create clear, labeled dialogue flows suitable for downstream analysis in Large Language Models.

• Multi-layered output: Generates transcription.txt (labeled), transcription-raw.txt (plain text), transcription-raw.json (word-level timings), and metadata.json.

• Standardized workspace integration: Automatically saves outputs to a structured directory structure at ~/Documents/transcripts/{category}/{title}-{date}/.

• Efficient workflow: Designed for high-volume ingestion of podcasts, interviews, video meetings, and lecture recordings.

• Usage Note: Always enclose URLs containing special characters like '&' in quotes to prevent command line execution errors.

• Dependencies: Requires yt-dlp and ffmpeg to be pre-installed via your system package manager (e.g., brew install yt-dlp ffmpeg).

• Configuration: Ensure that API credentials are properly set in the .env file located within the transcribee directory to enable diarization capabilities.

• Performance: Processing speed is subject to media length and system resource availability; utilize word-level timings from the raw JSON file for custom alignment tasks or building automated summary pipelines.