mozilla/DeepSpeech ? reverse-engineered prompt

Build me an offline speech to text engine that can run locally on a machine or small device, and can transcribe audio files from the command line and from Python. I want it to support using a trained model for inference, plus basic tools to evaluate accuracy and to help with training a new model from audio and transcripts. Please include simple example scripts, clear install and usage docs, and a way to package it so people can use it without needing a cloud service. If it helps, look up current docs online for anything unclear, but keep the experience focused on easy local transcription, model evaluation, and training. Make it practical for real world use on everything from a laptop to a Raspberry Pi style device.