Overview
It runs quantized multilingual models on ordinary CPUs under 2 GB of RAM, avoiding GPU hardware and recurring cloud speech fees.
Pricing
Free MIT-licensed code; model weights are downloaded from their publishers under their respective licenses.