rhasspy/piper ? reverse-engineered prompt

I want to build a local text-to-speech system that I can run on my own computer without needing an internet connection. It should be really fast and generate high-quality, natural-sounding voices using a neural network. I've heard about a project called Piper from Rhasspy that does this, so let's base it on that.

The main function should be simple: take some text as input and output an audio file, like a WAV. I'd also like the ability to switch between different voices, so please set it up to use a few of the pre-trained voice models that are available. It seems like the project might have moved to a new home on GitHub, so feel free to look up the latest version and documentation to make sure you're using the right source code. Please package it all up so it's easy to run, maybe with Docker.