µSpeech is best at detecting frictaves like s, z and f. It is particularly good at differentiating between a "s" and a "f" and one phoneme that is neither. A good vocabulary would use these letters in combination. For example if we have a car one could say the following:

Stop
Left
right

In this case we look for just the "f"s and the "s"s so our detector would sample the microphone, then use signal::getPhoneme() to get the s's and f's.