The approach can not only distinguish the voices in the room, but also the subjects. It's broad enough to both account for a sophisticated medical diagnosis and small talk like the weather. Doctors could have all the vital information they need for follow-ups and a better connection to their patients.
The system is far from perfect. The best voice recognition system in the study still had an error rate of 18.3 percent. That's good enough to be practical, according to the researchers, but it's not flawless. There's also the matter of making sure that any automated transcripts are truly private and secure. Patients in the study volunteered for recordings and will have their identifying information scrubbed out, but this would need to be highly streamlined (both through consent policies and automation) for it to be effective on a large scale.
If voice recognition does find its way into doctors' offices, though, it could dramatically increase the effectiveness of doctors. They could spend more time attending patients and less time with the overhead necessary to account for each visit. Ideally, this will also lead to doctors working more reasonable hours -- they won't burn out and risk affecting their judgment through fatigue.