TTS Engines

DPO Reader supports three text-to-speech backends with different trade-offs.

Bark (Default)

Neural TTS that runs locally. Produces natural-sounding speech with good intonation and emotion.

dpo-reader listen URL -e bark

• Quality: Excellent

• Speed: ~10 seconds per sentence

• Hardware: GPU recommended (MPS on Apple Silicon, CUDA on NVIDIA)

• Install: Included by default

Bark works without a GPU but runs slower. On Apple Silicon Macs, it uses Metal Performance Shaders automatically.

OpenAI

Cloud-based TTS with the best quality. Requires an API key and costs money per character.

export OPENAI_API_KEY=sk-...
dpo-reader listen URL -e openai

• Quality: Best

• Speed: Fast (network-dependent)

• Hardware: Any (runs in cloud)

• Cost: ~$0.015 per 1K characters

Get an API key at platform.openai.com/api-keys.

Available voices: alloy, echo, fable, onyx, nova, shimmer.

Piper

Lightweight local TTS optimized for CPU. Good for batch processing or machines without GPUs.

uv pip install dpo-reader[piper]
dpo-reader listen URL -e piper

• Quality: Good

• Speed: ~0.1 seconds per sentence

• Hardware: CPU only (~50MB models)

• Install: dpo-reader[piper]

Piper uses ONNX runtime and works on any machine. Models download automatically on first use.

Comparison

Engine	Quality	Speed	Requirements
OpenAI	Best	Fast	API key ($)
Bark	Excellent	Slow	GPU helps
Piper	Good	Fast	CPU only