Machine learning could turn a patient's voice into a diagnostic tool

It might be easy to diagnose a cold based on a patient's hoarse voice, but researchers believe subtle vocal changes undetectable to the human ear can identify or predict certain difficult-to-diagnose diseases. That's where machine learning could step in, helping physicians detect a range of illnesses that can't be diagnosed with more traditional tests, such as post-traumatic stress disorder, according to MIT Technology Review. Charles Marmar a psychiatrist at New York University's Langone Medical Center is collecting voice samples from combat veterans to determine how specific characteristics like tone and pitch could help psychiatrists diagnose PTSD, traumatic brain injury or depression. Initial studies have found vocal cues can differentiate PTSD patients from healthy ones with 77% accuracy, but an influx of new data will strengthen those results. "Medical and psychiatric diagnosis will be more accurate when we have access to large amounts of biological and psychological data, including speech features," Marmar told MIT Technology Review.