DPO Reader

Turn Discourse threads into podcasts. Each author gets their own voice.

I got tired of skimming long forum threads. DPO Reader converts them to audio so I can listen while doing other things. Point it at any Discourse thread and it'll synthesize the whole thing with different voices for each participant.

Install

# uv (recommended)
uv tool install dpo-reader

# Or run directly without installing
uvx dpo-reader listen "https://discuss.python.org/t/your-thread"

# pipx works too
pipx install dpo-reader

Usage

# Basic: generate audio and play it
dpo-reader listen "https://discuss.python.org/t/your-thread"

# Interactive TUI with read-along highlighting
dpo-reader listen "https://discuss.python.org/t/your-thread" --ui

# Start from a specific post (the /50 in the URL does this automatically)
dpo-reader listen "https://discuss.python.org/t/your-thread/12345/50" --ui

# Or explicitly with -s
dpo-reader listen "https://discuss.python.org/t/your-thread" -s 50

# Export without playing
dpo-reader export "https://discuss.python.org/t/your-thread" -o thread.wav

# See what you're getting into before generating
dpo-reader info "https://discuss.python.org/t/your-thread"
dpo-reader preview "https://discuss.python.org/t/your-thread"

Options

-o, --output PATH         Output file (default: output.wav)
-e, --engine ENGINE       openai | bark | piper
-s, --start-post INT      Start from this post number
-n, --max-posts INT       Limit number of posts
--ui                      Interactive TUI with controls
--no-attribution          Skip "Author says:" prefix
--no-play                 Don't auto-play after generating
-c, --cache-dir PATH      Cache audio chunks (useful if generation crashes)
-p, --pause FLOAT         Seconds between posts (default: 1.5)
-f, --file PATH           Load from local JSON (testing)

TTS Engines

Bark is the default. It runs locally and produces natural-sounding speech with good intonation. A GPU helps a lot but isn't strictly required.

OpenAI (-e openai) sounds the best if you have an API key. Get one at platform.openai.com/api-keys and set OPENAI_API_KEY in your environment or a .env file.

Piper (-e piper) is the lightweight option. Install with uv pip install dpo-reader[piper]. Good for batch processing or machines without GPUs.

TUI Controls

When using --ui:

Requirements

Python 3.10-3.14 (bark/torch). The optional piper extra doesn't support 3.14 yet: no onnxruntime release supports both 3.10 and 3.14, and torch has no 3.15 wheels.

Development

git clone https://github.com/JacobCoffee/dpo-reader.git
cd dpo-reader
make dev

make lint        # Linting
make type-check  # Type checking
make test        # Tests
make ci          # All of the above

# Test TTS locally
make test-listen  # Piper
make test-bark    # Bark

License

MIT