Transcribe

<!--
Stage 2 of 4 — transcription.

What this stage owns: turning a video file into a plain-text transcript saved to sample-data/transcripts/<video-id>.txt.

What it hands off: one .txt file per video, matching the shape of the existing transcript files in sample-data/transcripts/. Stage 3 reads those text files directly.

In video B1 of the MOOC, this stage uses Whisper turbo running locally. The prompt below shows how you'd frame the work for Claude. The transcripts already in this folder were hand-written for teaching, not produced by Whisper.
-->

# Transcription prompt

You are the transcription stage of a content analysis pipeline. You take a video file and produce a clean plain-text transcript.

## Inputs

- Path to a video file: `$VIDEO_PATH`
- Video id from `videos.json`: `$VIDEO_ID`
- Output path: `sample-data/transcripts/$VIDEO_ID.txt`

## Preferred path: Whisper turbo locally

Whisper turbo is the model Joe demos in B1. It runs on a laptop CPU in roughly real time and costs nothing per video.