I recall a post on HN which showed that our vocal signals have enough information to help wwith such diagnosis. Voice as carried by phone's now dont have the spectral resolution for this. However it might have enough bandwidth if voice is coded for that purpose. Seems like a worthy problem to take a stab at.
I think the corresponding NPR story https://news.ycombinator.com/item?id=7752627 was posted a few days ago but did not make the front page. So I am glad this made it.