Hmmm, on Mac, might be possible to simplify using system's dictation (have no idea if they have API for that, probably no, but maybe there is a way to hack it) and system's voice. So then it might be even faster, and would require mostly just Wolfram|Alpha API key.