Voice recognition question for next step.

Good afternoon, I'm developing a voice recognition to perform some tasks connected to an arduine board. Active already extracts as audio resources and I have already trained my neural network with 98% accuracy.

Doubt:

What would be the next step? How will I use a .mat generated by the best neural network training for the fact, classify me and pass on an Arduino board? Is there any project with the source code with the example? Thank you.