Well, I don't understand how it works well enough to know what feature to ask for...
I guess the request is that someone who knows how it works, check the effects on the voice recognition engine when you have a very small command database.
The reason I want to have few commands is because I always play music and that causes some problems with voice recognition. I'm already using prefix mode all the time, it's just annoyingly often random noise is interpreted as words with 90%+ confidence, including my prefix. I rarely reach that high when I actually say the same words...