Transcribe: Never Miss What Was Said

Turn YouTube videos and audio files into searchable, quotable transcripts.

The Problem

You consume valuable content through videos and audio, but:

• Can't search what was said - Information is locked in audio format
• Can't reference specific moments - No way to quote or cite exact timestamps
• Takes hours to review - Re-watching entire videos to find that one insight
• Miss key points - No written record means details get forgotten
• Can't share highlights - No easy way to extract and share the best parts

The Solution

Transcribe is a multi-stage pipeline that:

1. Downloads audio - From YouTube or uses your local files
2. Creates accurate transcripts - Using OpenAI's Whisper API
3. Formats for readability - Smart paragraphs with clickable timestamps
4. Extracts insights - AI-powered summaries and key quotes (optional)
5. Preserves everything - Keeps audio files for offline listening

The result: Searchable, readable transcripts with insights you can reference forever.

Quick Start

Prerequisites: Complete the Amplifier setup instructions first.

Basic Usage

# Transcribe a YouTube video
python -m scenarios.transcribe "https://youtube.com/watch?v=..."

# Transcribe a local audio file
python -m scenarios.transcribe podcast.mp3

The tool will:

1. Download/extract the audio
2. Send it to Whisper API for transcription
3. Format into readable paragraphs
4. Generate AI insights (summary + quotes)
5. Save everything in organized folders

Your First Transcript

1. Find a video to transcribe

Choose a YouTube video or prepare an audio file:

• YouTube: Copy the video URL
• Local file: Note the path to your .mp3, .mp4, .wav, etc.

2. Run the transcription

python -m scenarios.transcribe "https://youtube.com/watch?v=dQw4w9WgXcQ"

3. Watch the progress

The tool will show:

Downloading audio from YouTube...
Extracting audio (this may take a moment)...
Transcribing with Whisper API...
Formatting transcript into readable paragraphs...
Generating AI insights...
✓ Transcript saved to: ~/amplifier/transcripts/dQw4w9WgXcQ/

4. Explore the output

Navigate to your transcripts folder:

~/amplifier/transcripts/dQw4w9WgXcQ/
├── audio.mp3          # The audio file (for offline listening)
├── transcript.md      # Readable transcript with timestamps
└── insights.md        # AI summary and key quotes

5. Use your transcript

• Read the formatted transcript with clickable timestamps
• Search for specific topics or quotes
• Listen to the preserved audio file
• Share insights and quotes with proper citations
• Reference exact moments with timestamp links

Usage Examples

Basic: Single YouTube Video

python -m scenarios.transcribe "https://youtube.com/watch?v=..."

What happens:

• Downloads audio from YouTube
• Creates accurate transcript
• Formats into readable paragraphs
• Generates summary and quotes
• Saves everything for future reference

Advanced: Multiple Sources

python -m scenarios.transcribe video1.mp4 "https://youtube.com/..." podcast.mp3

What happens:

• Processes each source sequentially
• Saves state between each item
• Creates separate folders for each
• Updates the transcript index
• Can resume if interrupted

Resume Interrupted Session

# If interrupted, just add --resume
python -m scenarios.transcribe --resume video1.mp4 video2.mp4

What happens:

• Finds where you left off
• Skips already completed items
• Continues from interruption point
• Preserves all previous work

How It Works

The Pipeline

YouTube URL or Audio File
         ↓
    [Download/Extract Audio]
         ↓
    [Whisper Transcription]
         ↓
    [Format Paragraphs] ────→ transcript.md
         ↓
    [Generate Summary] ─────→ insights.md
         ↓
    [Update Index] ─────────→ index.md

Key Components

• Video Loader: Downloads from YouTube using yt-dlp
• Audio Extractor: Compresses audio for API limits (25MB max)
• Whisper Transcriber: Calls OpenAI's speech-to-text API
• Transcript Formatter: Creates readable paragraphs with timestamps
• Insights Generator: AI summaries and quote extraction
• State Manager: Enables interrupt/resume capability

Why It Works

Code handles the structure:

• Audio download and extraction
• API calls and retry logic
• File organization and caching
• State management for resume
• Error handling and recovery

AI handles the intelligence:

• Accurate speech transcription
• Summary generation
• Key quote identification
• Content understanding

This separation means reliable processing (code) with intelligent output (AI).

Configuration

Output Locations

User Content (~/amplifier/transcripts/):

• index.md - Auto-generated index of all transcripts
• [video-id]/audio.mp3 - Preserved audio file
• [video-id]/transcript.md - Readable transcript
• [video-id]/insights.md - Summary and quotes

Technical Artifacts (.data/transcripts/):

• [video-id]/transcript.json - Structured data
• [video-id]/transcript.vtt - WebVTT subtitles
• [video-id]/transcript.srt - SRT subtitles

Audio Caching

Audio files are automatically cached to save bandwidth:

• First run downloads and saves audio
• Subsequent runs use cached version
• Force re-download with --force-download

Cost Estimation

OpenAI Whisper API pricing (as of 2024):

• $0.006 per minute of audio
• Example: 60-minute video = $0.36
• Cost shown before processing

Troubleshooting

"yt-dlp is not installed"

Problem: Missing YouTube download dependency.

Solution:

make install  # or: uv add yt-dlp

"ffmpeg not found"

Problem: Audio processing tool not installed.

Solution:

• macOS: brew install ffmpeg
• Ubuntu: sudo apt-get install ffmpeg
• Windows: Download from ffmpeg.org

"Audio file too large"

Problem: File exceeds 25MB API limit.

Solution: The tool auto-compresses. If it still fails, manually compress:

ffmpeg -i input.wav -b:a 64k -ar 16000 output.mp3

"API key not found"

Problem: OpenAI/Anthropic API keys not configured.

Solution: Set in .env file:

OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...

Learn More

• HOW_TO_CREATE_YOUR_OWN.md - Create your own tool like this
• Amplifier - The framework that powers these tools
• Scenario Tools - More tools like this one

What's Next?

This tool demonstrates what's possible when you describe a process to Amplifier:

1. Use it - Transcribe videos and build your knowledge library
2. Learn from it - See HOW_TO_CREATE_YOUR_OWN.md for how it was made
3. Build your own - Describe your goal to Amplifier
4. Share back - Let others learn from what you create!

---

Built through conversation using Amplifier - The entire tool came from describing the goal in natural language. See HOW_TO_CREATE_YOUR_OWN.md for the actual conversation.