A while back, somebody (not me) posted a link here to a hobby project I did:
"Simple Speech-to-Text on the '10 Cents' CH32V003 Micro-controller"
https://news.ycombinator.com/item?id=40504481
So I though I'd submit a link to a follow-up project, that uses an 8-bit 'Arduino Nano' micro-controller instead to see how low-power I can go doing real-time MFCC-based feature extraction for spoken audio samples. Because reasons. (proof-of-concept more than anything).