youtube-transcription

name	YouTube Transcription
description	Transcribe a YouTube video or local audio file using AssemblyAI (paid, speech-to-text AI). Choose this skill over youtube-to-markdown when the video has no auto-generated captions, is in a non-English language that needs accurate transcription, or you need a local audio file transcribed. Use when the user says "transcribe this video with AssemblyAI", "get speech-to-text transcript", "transcribe this MP3/WAV/M4A", or "transcribe a video that has no subtitles". Requires ASSEMBLYAI_API_KEY.
version	0.1.0

YouTube Transcription Skill

Convert YouTube videos and audio files to searchable, formatted markdown transcripts using AssemblyAI.

When to Use This Skill

Use this skill when you need to:

• Transcribe YouTube videos into markdown documents
• Convert audio files (MP3, WAV, M4A) to text transcripts
• Create searchable transcripts with metadata and proper formatting
• Add transcripts to knowledge bases or project documentation

Core Workflow

The skill provides a three-step workflow:

1. Download Audio (if needed) - Extract audio from YouTube video using yt-dlp with Firefox cookie authentication
2. Transcribe - Send audio to AssemblyAI API for speech-to-text conversion
3. Format & Save - Create markdown document with metadata and transcript text

Prerequisites

• AssemblyAI API key - Free account at https://www.assemblyai.com (set as ASSEMBLYAI_API_KEY environment variable)
• Firefox browser - For YouTube authentication (only needed for video downloads)
• yt-dlp - Automatically installed by uv run

Quick Start

From YouTube URL

export ASSEMBLYAI_API_KEY="your_key_here"
cd /path/to/vault
uv run --with=assemblyai ./.claude/skills/youtube-transcription/scripts/transcribe_video.py "https://youtu.be/VIDEO_ID"

From Local Audio File

export ASSEMBLYAI_API_KEY="your_key_here"
uv run --with=assemblyai ./.claude/skills/youtube-transcription/scripts/transcribe_video.py /path/to/audio.mp3

Output: Full transcript printed to stdout + saved to /tmp/transcript.txt

Common Tasks

Task 1: Transcribe a YouTube Video and Save to Project

# 1. Run transcription
export ASSEMBLYAI_API_KEY="your_key_here"
uv run --with=assemblyai ./.claude/skills/youtube-transcription/scripts/transcribe_video.py "https://youtu.be/_gPODg6br5w"

# 2. Copy output to project with metadata
# Edit the transcript file to add:
# - Title
# - Video ID, URL, Date
# - Proper markdown formatting

# 3. Save to your project
cp /tmp/transcript.txt pages/Projects/My-Project/Video-Title-Transcript.md

Task 2: Download YouTube Audio First (if video requires authentication)

# Download with Firefox cookies
yt-dlp -x --audio-format mp3 --cookies-from-browser firefox -o "video.mp3" "https://youtu.be/VIDEO_ID"

# Then transcribe the saved file
export ASSEMBLYAI_API_KEY="your_key_here"
uv run --with=assemblyai ./.claude/skills/youtube-transcription/scripts/transcribe_video.py ./video.mp3

Task 3: Batch Transcribe Multiple Videos

Create a simple script to loop through videos:

#!/bin/bash
export ASSEMBLYAI_API_KEY="your_key_here"
SKILL_PATH="./.claude/skills/youtube-transcription/scripts/transcribe_video.py"

for url in \
    "https://youtu.be/VIDEO1" \
    "https://youtu.be/VIDEO2" \
    "https://youtu.be/VIDEO3"
do
    echo "Transcribing: $url"
    uv run --with=assemblyai $SKILL_PATH "$url"
    sleep 5  # Rate limiting
done

Troubleshooting

Problem: "Sign in to confirm you're not a bot"

YouTube is blocking yt-dlp. This happens with age-restricted or new videos.

Solution: Use Firefox cookies with JavaScript runtime:

yt-dlp -x --audio-format mp3 \
    --cookies-from-browser firefox \
    --js-runtimes bun \
    --remote-components ejs:github \
    -o "video.mp3" \
    "https://youtu.be/VIDEO_ID"

Problem: AssemblyAI module not found

Script needs the AssemblyAI package at runtime.

Solution: Always use --with=assemblyai flag:

uv run --with=assemblyai ./scripts/transcribe_video.py <input>

Problem: API key not recognized

AssemblyAI requires valid API key.

Solution: Verify environment variable is set:

echo $ASSEMBLYAI_API_KEY  # Should show your key

Set it in your shell profile to persist:

# Add to ~/.bashrc or ~/.zshrc
export ASSEMBLYAI_API_KEY="your_actual_key_here"

Output Format

The skill outputs raw transcript text. Format it as markdown with metadata:

# Video Title

**Source:** YouTube - Creator Name (@handle)
**Video ID:** VIDEO_ID
**URL:** https://youtu.be/VIDEO_ID
**Date:** YYYY-MM-DD
**Duration:** MM:SS

---

## Full Transcript

[Transcript text...]

---

## Key Sections

- [Timestamp] Section title
- [Timestamp] Another section

Integration Tips

For [[Second Brain]] Project

Save transcripts following this structure:

pages/Projects/Second Brain/
├── Nate B Jones - Why 2026 Is the Year to Build a Second Brain - Transcript.md
├── Nate B Jones - Follow-up Video - Transcript.md
└── README.md (links to transcripts)

Update the project README to include:

- `[[Title - Transcript]]` - Full video transcript

For Knowledge Base

Create a transcripts folder:

pages/PKB/Video Transcripts/
├── Topic 1/
│   └── Video Title - Transcript.md
├── Topic 2/
│   └── Another Video - Transcript.md

Cost & Limits

• AssemblyAI pricing: ~$0.60-1.00 per hour of audio
• Free tier: Limited minutes per month
• Check usage: https://www.assemblyai.com/dashboard
• Language support: English (best), others supported but less accurate

Advanced Options

See references/assemblyai_options.md for:

• Language detection
• Speaker diarization (multiple speakers)
• Entity detection
• Summarization options
• Custom punctuation rules

Related Skills

• [[Second Brain]] - Knowledge management project using transcripts
• [[Git Workflow]] - For committing transcripts to version control

Resources

• AssemblyAI documentation: https://www.assemblyai.com/docs/
• yt-dlp documentation: https://github.com/yt-dlp/yt-dlp/wiki
• Skill implementation: .claude/skills/youtube-transcription/scripts/transcribe_video.py