transcribe

The transcribe skill converts audio files to text using OpenAI's transcription models. It solves the problem of manually transcribing speech from recordings by automating the process through a bundled CLI script (`transcribe_diarize.py`). The skill handles everything from simple text extraction to complex multi-speaker scenarios, and is designed to be deterministic and repeatable through its CLI-first approach.

The skill offers two primary modes: fast plain-text transcription using `gpt-4o-mini-transcribe` for quick results, and speaker-diarized transcription using `gpt-4o-transcribe-diarize` which labels who said what in a conversation. It supports known-speaker hints — up to 4 reference audio samples — so it can identify specific people by name rather than generic labels. Long audio is handled automatically via chunking, and outputs can be organized into job-specific directories to avoid overwriting previous results. Supported output formats include plain text and structured JSON.

This skill is well-suited for journalists transcribing interviews, teams documenting meeting recordings, researchers processing oral history audio, and developers building transcription pipelines. Anyone working with audio or video content who needs accurate, labeled text output will benefit from this skill. It requires an OpenAI API key and Python with the `openai` package installed, making it accessible to users with basic technical familiarity.