Cognitive disorders can be particularly debilitating, costing $200 billion in the USA alone. Clinical interviews, laboratory tests, and imaging scans may used to evaluate cognitive health, but are costly to acquire and process. This motivates the exploration of measures that are easier to record and are less invasive, specifically, speech.
In collaboration with Prof. Rhoda Au and the Framingham Heart Study, we extract and identify audio and text features that are most predictive of cognitive impairment from audio recordings of 5,000 subjects undergoing neuropsychological evaluations (composed of memory, recall, and comprehension tests), collected over the past 10 years. We also look to detect depression from speech.