Among all the options I had to recognize the user voice commands I choosed the Watson Speech to Text service. This choice was based on how simple is to use the service after creating an evaluation account. Also, they provice simple examples of API calls using NodeJS.
I was wondering how to keep a constant open connection with the service and receive events when specific keywords were spoken, i.e. UP or DOWN. Instead of programming everything from scratch I used the TJBOT lib, a NodeJS project that was used for the IBM's TJBot cardbord robot. This allow me to have a prototype with very few lines of code. Here's the link for it: