transcribe-youtube

name	transcribe-youtube
description	Download and transcribe YouTube videos to markdown. Use when user says "transcribe youtube", "transcribe this video", "youtube transcript", or provides a YouTube URL to transcribe.

YouTube Video Transcription

This skill downloads YouTube videos and transcribes them using Whisper, outputting a markdown file to the Obsidian vault.

Configuration

• Output directory: ~/obsidian/Intake/
• Temp directory: /tmp/
• Whisper model: medium (good balance of speed and accuracy)

Workflow

Step 1: Extract Video ID and Metadata

First, get the video metadata:

yt-dlp --print "%(id)s|||%(title)s|||%(channel)s|||%(upload_date)s" "<URL>"

Parse the output to extract:

• id - YouTube video ID
• title - Video title
• channel - Channel name
• upload_date - Upload date (YYYYMMDD format)

Step 2: Download Audio

Download only the audio (much faster than full video):

yt-dlp -x --audio-format mp3 -o "/tmp/%(id)s.%(ext)s" "<URL>"

This creates /tmp/<video-id>.mp3

Step 3: Transcribe with Whisper

Run Whisper transcription:

whisper /tmp/<video-id>.mp3 --model medium --language en --output_format txt --output_dir /tmp

This creates /tmp/<video-id>.txt with the transcript.

Step 4: Create Markdown File

Read the transcript and format as markdown with frontmatter:

---
title: "<Video Title>"
source: <YouTube URL>
channel: <Channel Name>
date: <YYYY-MM-DD>
transcribed: <today's date>
type: transcript
---

# <Video Title>

**Source:** [<Channel Name>](<YouTube URL>)
**Date:** <YYYY-MM-DD>

---

<Transcript content>

Step 5: Save to Intake

Sanitize the title for use as filename (remove special characters, limit length):

• Replace /, \, :, *, ?, ", <, >, | with -
• Limit to 100 characters
• Trim whitespace

Save to: ~/obsidian/Intake/<Sanitized-Title>.md

Step 6: Cleanup

Remove temporary files:

rm /tmp/<video-id>.mp3 /tmp/<video-id>.txt

Step 7: Confirm

Tell the user:

• The file was created at Intake/<filename>.md
• Video title and channel
• Approximate transcript length

Error Handling

• Invalid URL: If yt-dlp fails to recognize the URL, inform the user it's not a valid YouTube URL
• Download failed: Check if video is private, age-restricted, or region-locked
• Whisper fails: Check if audio file exists and is not corrupted
• Long videos: Warn user that videos over 1 hour may take several minutes to transcribe

Notes

• Only downloads audio to save time and bandwidth
• Uses Whisper medium model for good accuracy without excessive processing time
• Transcripts go to Intake/ for later processing into Zettelkasten notes