Audio version — Estimated duration: 2 min 45 sec

Requirements

RequirementMinimum
Python3.10+
Disk~500 MB for Piper models

Online mode (default)

Minimal install — just the APIs:

pip install edge-tts openai

Set your Cerebras API key:

$env:CEREBRAS_API_KEY = "your-key-here"

Local TTS mode (Piper + GPU)

Add onnxruntime and the Piper wrapper:

pip install piper-tts onnxruntime-gpu huggingface-hub mutagen

onnxruntime-gpu includes CPU fallback. You do not need a separate onnxruntime install.

Piper models auto-download to models/ on first use from Hugging Face.

Local LLM mode (llama.cpp)

No extra pip packages. Run a llama.cpp server:

llama-server -m your-model.gguf --port 8080

The pipeline expects it at http://localhost:8080/v1.

For GPU acceleration on local TTS:

pip install onnxruntime-gpu

Force CPU fallback at runtime with --cpu.


Next: Quickstart