Old Microsoft Speech API would be a good fit here. I miss it. Back in 2007 I made myself a voice control interface for changing music playback. Trainable, completely off-line. Worked like a charm.