- Depression is significantly underdiagnosed.
- Research has identified vocal biomarkers of depression.
- Voice analysis apps might help diagnose and monitor depression.
It has been estimated that over a quarter of adults in the U.S. have some sort of mental health disorder, but less than half of those adults seek treatment in any given year. This lack of care is due, in part, to problems with access, but also to the social stigma that many feel in seeking help.
As well, many patients and health care providers don’t recognize the physical signs of depression, such as changes in facial expressions and changes in voice. But what if there was an easy, widely available, and private way to use such biomarkers to assess someone’s mental health status?
Welcome to the new age of digital medicine. Recent technological advancements now make it possible to analyze simple voice recordings using machine learning techniques to both screen for and monitor depression. Armed with only a smartphone, a device found widely even in areas lacking much in the way of health care, and a quick voice sample, mental health professions can take advantage of a noninvasive way to assess a patient’s mental health.
Using AI to Detect Vocal Biomarkers
The use of voice analysis to help monitor illness is a growing area of applied health-based artificial intelligence (AI) research. Diagnostic approaches that make use of similar machine learning and classification tasks have begun being used in medical contexts, such as to screen for COVID or to assess flare-ups of COPD.
These advances are possible because of subtle but measurable changes in the acoustics of speakers’ voices when illness-related respiratory symptoms affect the sound of their voice. But how we sound is not only a result of physical characteristics; it is also the result of a speaker’s cognitive processing and motor control.
Emotion research has found that voice-related measures are associated with different emotions. Speech rate, voice quality, pitch, and speech intensity have all been shown to be sensitive to speaker emotional state. For instance, depressed speakers have been found to have a slower speaking rate and less pitch variation, speak more softly and show reduced speech-motor coordination.
Alone these features may be subtle and not easy for a therapist to notice, much less find predictive of depression. But harnessing the power of computational algorithms, new research suggests that analyzing a complex of voice features might provide a cost-effective and simple way to diagnose and monitor depression.
In studies designed to determine how successfully voice feature extraction and analysis could identify depressed individuals, researchers needed only a short voice sample, collected by having participants answer a simple question like “What is your age?” or make a sustained vowel sound like “aahhhh” into a smartphone app.
Participants also filled out demographic and health questionnaires. The researchers then used the recordings as training and test data for depression detection models. The result? In most cases, the accuracy of the models was about 75 to 80% in evaluating depression from analysis of voice features. Several AI companies, such as Sonde Health and CompanionMx, have already started working with health care providers to bring these types of tools into clinical settings.
Quick, Private, and Accessible
While this might not be the 100 percent accuracy rate we would aim for ideally, it quickly automates the detection process using a widely available device (e.g., a smartphone) and does not require a patient to be present physically. This is a great option for patients who hesitate to come into an office due to social stigma or because of a lack of transportation or proximity, though of course would require a patient be aware of and consent to the process. Another benefit is that it could help identify patients suffering from depression during routine doctor visits, especially those who might feel uncomfortable directly talking about their mental health with the doctor.
Even more promising, it could be a tool not just to initially screen a patient but also to serve as an early warning system that a patient might be in need of help or a change in treatment plan. In fact, some of the apps already developed prompt users (again with consent) to record regular short voice samples that can be analyzed to check for biomarkers of depression and provide their health care teams a way to better track ups and downs between visits.
In short, our smartphones may very well be the next frontier in mental health treatment—allowing more of those that suffer from depression unobtrusive options for screening and monitoring. The key, of course, is that such technology, which just blindly detects patterns, be used in concert with health professionals, and not in lieu of them.
LinkedIn image: SeventyFour/Shutterstock. Facebook image: GaudiLab/Shutterstock.
Mundt, J. C., Vogel, A. P., Feltner, D. E., & Lenderking, W. R. (2012). Vocal acoustic biomarkers of depression severity and treatment response. Biological psychiatry, 72(7), 580–587.
Pan W, Flint J, Shenhav L, Liu T, Liu M, Hu B, et al. (2019) Re-examining the robustness of voice features in predicting depression: Compared with baseline of confounders. PLoS ONE 14(6): e0218172.
Ozkanca, Y., Öztürk, M. G., Ekmekci, M. N., Atkins, D. C., Demiroglu, C., & Ghomi, R. H. (2019). Depression Screening from Voice Samples of Patients Affected by Parkinson's Disease. Digital biomarkers, 3(2), 72–82.