Good question. Don’t have a great answer necessarily. Currently work on CS is background (I have a day job and CS consulting work). One underlying long term goal is that CS be the dominant platform for NLP toolkits; that it becomes more useful and more widely used than the Perl NLP Toolkit or the Stanford tools (parser postagger,etc). That requires brining the english pos/parser of CS to be more accurate than the gold standards of statistical parsing and tagging. CS is already faster than Stanford and can fit on a cell phone. But fitting on a cell phone is part of the ongoing requirements for CS.
I’m not current in the deep learning space, and don’t see it as getting voice-activated UI working as fast as just scripting it. I do see CS providing more support for speech recognition.