I had this idea on a voice interface for JOSM and created a small prototype called speech2JOSM.
It’s nothing big and still WIP, but maybe you want to give it a try to explore this kind of feature.
If there is a demand for the an better implementation, I’m looking for somebody who like to port it as native JOSM Java plugin? I guess this makes sense, if the detection rate get’s better…
Oh and maybe you would like to improve the speech to text detection, by supporting the crowdsourced recordings at voice.mozilla.org 
?