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?