Basic voice recognition modules need to be trained in the exact environment you're using them in.
That includes board position (including angle), speaker<>mic distance, user intonation, room reverb, etc.
Speaking louder when it doesn't work, doesn't work. Unless you have trained it for that.
A 1meter distance might work better than a 25cm distance.
Experiment with different activation commands. Hicell might not be that easy to 'understand'.
A three syllable word with distinct/different sounds might work better.
It took me several weeks of experimenting with an EasyVR module to get to an acceptable 90% success rate.
Leo..