Podcast Transcription Setup

Ultra Quick Start

1. Install uv (if you don’t have it):

   curl -LsSf https://astral.sh/uv/install.sh | sh
   

2. Run the self-contained script:

   ./transcribe_episodes.py
   

That’s it! The script automatically manages its own dependencies with uv.

What It Does

• Batch processes all 9 episodes from your MP3s/ directory
• Uses faster-whisper with large-v3 model - 4-5x faster than original Whisper
• Lower memory usage - ~2GB vs ~6GB with original whisper
• Voice Activity Detection - Better handling of silence and music
• Creates multiple output formats:
  • .txt - Clean plain text transcript
  • .srt - Subtitle file with timestamps (for video)
  • .vtt - Web-compatible subtitles
  • .json - Detailed data with word-level timing

Output Structure

transcripts/
├── 01-rizzoli-matt_rodbard.txt
├── 01-rizzoli-matt_rodbard.srt  
├── 01-rizzoli-matt_rodbard.json
├── 02-notion-rob_giampietro.txt
├── ...etc

Performance Notes

• First run: Downloads faster-whisper model (~3GB for large-v3)
• Processing time: ~2-5% of audio length (1-3 minutes per 60-minute episode)
• Memory usage: ~2GB (much lower than original whisper)
• Accuracy: Same as original Whisper, with better VAD (Voice Activity Detection)
• Language: Set to English with confidence scoring

Customization Options

Use faster model (lower quality):

model = whisper.load_model("base")  # Much faster

Add speaker detection:

# In transcribe_episode function, add:
result = model.transcribe(mp3_path, 
                        language='en',
                        word_timestamps=True,
                        verbose=False,
                        initial_prompt="This is a podcast conversation between Craig Mod and a guest.")

Output only specific formats: Comment out the formats you don’t want in the save_transcript() function.

Troubleshooting

“No module named ‘whisper’”:

pip install openai-whisper

Memory errors: Use smaller model:

model = whisper.load_model("base")

FFmpeg errors:

# macOS
brew install ffmpeg

# Ubuntu/Debian  
sudo apt install ffmpeg

Very slow performance: Consider using faster-whisper:

pip install faster-whisper

Integration Ideas

Once you have transcripts, you could:

1. Add to episode pages - Include transcript sections
2. Enable search - Make episodes searchable by content
3. Generate show notes - Extract key topics automatically
4. Create clips - Find quotable moments with timestamps

Craig Mod is a writer and photographer based in Japan. He's the author of the books Things Become Other Things and Kissa by Kissa and is a MacDowell, Ragdale, and VCCA writing fellow. His essays and articles have appeared in The New York Times, Eater, The Atlantic, California Sunday Magazine, Wired, Aeon, New Scientist, Virginia Quarterly Review, The New Yorker, The Morning News, Codex: Journal of Typography, and elsewhere.

He writes newsletters, oh yes, newsletters: Roden & Ridgeline.

The work on this site is supported 100% by paid memberships and book & print sales.

Whatever you do, don't follow @craigmod on Bluesky or Instagram.