Podscript

Podcast → Transcript!

Transcribe any podcast episode or YouTube video from the command line. Generates clean markdown with speaker diarization and timestamps. Works with the ElevenLabs API or fully locally using Whisper - no API key required.

Installation

# Local transcription (free, no API key needed)
pip install podscript[local]

# Or use ElevenLabs API
pip install podscript
podscript --setup  # paste your ElevenLabs API key

For local mode, just add --local to any command. For ElevenLabs, you'll need an API key.

For YouTube support, also install yt-dlp and ffmpeg.

Usage

# Transcribe a podcast from an Apple Podcasts link
podscript "https://podcasts.apple.com/us/podcast/huberman-lab/id1545953110?i=1000690"

# Transcribe a YouTube video
podscript "https://www.youtube.com/watch?v=dQw4w9WgXcQ"

# Use an RSS feed directly
podscript https://feeds.simplecast.com/JGE3yC0V

# Browse episodes first
podscript https://feeds.simplecast.com/JGE3yC0V --list

# Search for a specific episode
podscript https://feeds.simplecast.com/JGE3yC0V --search "AI"

# Pick episode #3 from the list
podscript https://feeds.simplecast.com/JGE3yC0V --episode 3

# Custom output filename
podscript https://feeds.simplecast.com/JGE3yC0V --latest --output transcript.md

Without any flags, the default behavior is to transcribe the most recent episode.

Output

Generates a markdown file with speaker labels and timestamps:

# The Economics of Carbon Removal

**Podcast:** a16z Podcast
**Date:** 2/10/2026
**Duration:** 1:04:23

---

## Speaker 1
[0:00] Welcome back to the show. Today we're talking about...

## Speaker 2
[0:15] Thanks for having me. So the key challenge with carbon removal is...

## Speaker 1
[2:41] That's fascinating. How does the economics actually work at scale?

Local Transcription

You can transcribe entirely offline using a local Whisper model — no API key required:

pip install podscript[local]

This installs faster-whisper, pyannote.audio, and torch.

Usage

# Basic local transcription (uses "base" model, no speaker diarization)
podscript "https://www.youtube.com/watch?v=..." --local

# Use a larger model for better accuracy
podscript "https://www.youtube.com/watch?v=..." --local --model medium

# Enable speaker diarization with a HuggingFace token
podscript "https://feeds.example.com/rss" --local --hf-token hf_xxxxx

# Or set the token as an environment variable once
export HF_TOKEN=hf_xxxxx
podscript "https://feeds.example.com/rss" --local

Model Sizes

CPU mode uses int8 quantization automatically. GPU (CUDA) uses float16.

Speaker Diarization

Speaker diarization (identifying who said what) requires a free HuggingFace token:

1. Create an account at huggingface.co
2. Accept the terms for pyannote/speaker-diarization-3.1
3. Create a token at huggingface.co/settings/tokens
4. Pass it via --hf-token or set HF_TOKEN in your environment

Without a token, all speech is attributed to "Speaker 1" — still useful for single-speaker content.

License

MIT