How It Works

The Process

We take influential machine learning and systems papers and transform them into ~25-30 minute podcast episodes. Each episode features a conversational format with two hosts—Alex and Maya—who break down complex technical concepts into accessible explanations.

What Each Episode Covers

• The core problem and motivation behind the research
• Key technical contributions and innovations
• Practical implications and real-world usage
• Comparisons with alternative approaches
• Best practices and recommendations

Episode Format

Each paper folder in our repository contains:

Technology

Our podcasts are generated using ElevenLabs text-to-speech with their turbo model for realistic, natural-sounding voices. We use two distinct voices for the hosts to create an engaging conversational dynamic.

Voice Configuration

• Alex — Adam voice (deep male, British accent)
• Maya — Rachel voice (clear female, American accent)

Regenerating Episodes

Each episode can be regenerated using the included Python script. You'll need an ElevenLabs API key and ffmpeg installed.

# Set your API key
export ELEVENLABS_API_KEY="your-api-key"

# Install dependencies
pip install elevenlabs

# Generate the podcast
cd zhao-2023-pytorch-fsdp
python generate.py

Contributing

Want to add a paper? The script format is straightforward—check out any existing script.md file for the expected structure. The key elements are:

• Speaker lines prefixed with **ALEX:** or **MAYA:**
• Section headers with timestamps (used for organization)
• Natural, conversational language that explains technical concepts

Listening Tips