It’s great to hear that you’re interested in experimenting with the keyword spotter using real-time audio input. It is definitely possible to connect the GUI to an microphone to capture realtime audio and pass it to a Nengo model. Roughly, you’ll need to (1) buffer the incoming audio signal, (2) compute standard audio features (MFFCs) on the buffered signal, and (3) stream these features into the model.
We do have a preliminary GUI interface doing these things, but it is in an unsupported and unreleased state. We’re looking at making it publicly available to experiment with on a “no-guarantees” basis, so I’ll post back here shortly once I’ve got more information for you on this front.
Let me know if you have any other questions!