A software solution, specially designed for the early detection of Post-Traumatic Stress Disorder (PSTD) is currently under study at the New York University (NYU) School of Medicine. Using voice analysis, the artificial intelligence can determine with eighty-nine percent (89 %) accuracy if a person is suffering from the stress disorder.
The author of the study is Dr. Charles R. Marmar, MD, head of the Department of Psychiatry and the Lucius N. Littauer Professor at the university’s School of Medicine. He reports that their findings indicate that with further validation and refinement of the program, clinics can use it to immediately diagnose whether or not a patient has the disease.
Dr. Marmar explained that those suffering from PTSD tend to talk slower, sound monotonous and have rare bursts of vocalization. When suffering from the disorder, they are less energetic, seemingly lifeless with their speech, using a flatter tone and show longer hesitation to speak up.
A report of this study was published online in the “Depression and Anxiety” journal last April 22, 2019, which provided some details of how Dr. Marmar and the NYU team proceeded in testing the accuracy of the PTSD diagnostic program.
How the PTSD Detection Software Analyzes by Way of Voice
The study’s researchers acknowledged that PTSD diagnosis performed by way of clinical interview or based on a self-report evaluation, are intrinsically prone to bias. Thus the need to develop a more objective method of determining and measuring the progression of PTSD by way of physical markers; whilst using similar laboratory values taken into account when diagnosing medical conditions.
Gathering Clinician-Administered PTSD Scale (CAPS), which are hour-long diagnostic interviews, the researchers recorded the CAPS of veterans with PTSD and without PTSD. The voice recordings of 53 Afghanistan and Iraqi veterans with stress disorder, and of 78 veterans without PTSD, were fed into a voice software furnished by SRI International (formerly Standford Research Institute). The recordings yielded 40,526 speech-based traits that came out as brief spurts of talk, from which the PTSD program research teams combed through to establish speech patterns.
In the their study, the PTSD research team applied a statistical machine-learning program called Random Forests, The latter is an AI built with decision rules and algorithms, making it capable of knowing how to classify individuals based on examples. The greater the amount of training data built into the AI, the higher the level of accuracy achieved by its decision-making processes.
The Random Forest then linked the voice characteristics recognized as symptoms of PTSD, including the standard indications such as less clear speech and flat lifeless tone, to the patterns specific to the voice features of those with PTSD. The current study though, did not delve on the mechanisms causal to PTSD. Mainly because the theory, is that traumatic experiences tend to alter brain circuits involved in processing emotion and muscle tone, which as a result, affects the voice of the afflicted person.
The Director of Speech Technology and Research (STAR) Laboratory at SRI, Dimitra Vergyn, remarked that the speech analysis software used in the PTSD detection, is one of several capabilities included in SRI’s SenSay Analysis, a speech analytics platform. This technology takes into account the tone, the rhythm, the frequency and articulatory traits of speech. At the same time, it also analyzes words in assessing the physical, emotional and mental health of the speaker, as well as the quality of cognition, sentiment and communication manifested.