Onboard AI voice recognition

I'm trying to build a specialized voice recognition inference model using Edge Impulse tools, and so far I got interesting but not good enough results that are not worth releasing the code.

The model seems to be doing pretty well when running on my PC, recognizing "open" and "close" commands, but very poorly once loaded on the XIAO ESP32-S3.

I'll eventually train it using only my voice, but that requires recording at least 15 minutes of each word, in different environments, which is a bit tedious. So for now I'm mixing samples of my voice and OpenAI generated samples, along with "noise" and "random" datasets from Edge Impulse.