I want to train my own commands

I already have the model with my own recorded audio, but it's too big. I don't know how to adjust the size. I have four commands in Spanish and a noise folder. What can I do to implement my own commands in the code?

No file, test, or entry point is named. Start by locating the training notebook and the ESP32 TensorFlow Lite integration, then determine how the four Spanish commands and noise data are represented and what model-size limit must be met. Done means a reproducible path for training and integrating custom commands.

Written by the indexing model from the issue text.