What problem does it solve? Converting recorded lectures and audio files into searchable Korean text is slow and error-prone when done manually, and local CPU transcription of long recordings can take hours. This Skill automates speech-to-text transcription and the entire post-processing pipeline so transcripts become split, indexed, and progress-tracked study material. ## Core Features & Use Cases - Colab GPU Transcription: Runs faster-whisper large-v3 on a free T4 GPU via a Colab notebook, transcribing a 1.5-hour lecture in about 5 minutes. - Local Transcription: Uses whisper.cpp (ggml-small model) through transcribe.bat for single files or transcribe_batch.bat for whole folders, producing .txt output next to the source audio. - Automated Post-Processing: post_transcribe.py detects new transcripts, splits them into parts, re-indexes the qmd law-notes vector collection, updates study progress, and archives processed files with a JSON transcription log. - NotebookLM Extraction: extract_notebooklm.py pulls the main Korean transcript text out of saved NotebookLM HTML files. - Use Case: Upload a recorded civil law lecture to the inbox folder, run the Colab notebook, then execute the post-processing script to get a split, vector-indexed transcript with progress tracking updated automatically. ## Quick Start Upload your audio file to the 녹음 inbox folder and ask the assistant to transcribe the lecture and run the post-processing pipeline.
