AI detects depression from speech
Register for free and continue reading
Join our growing army of changemakers and get unlimited access to our premium content
Deep learning can spot depression based on speech and brain activity
Spotted: Globally, approximately 280 million people suffer from depression, and although there are effective ways to treat the illness, unless a person seeks help, the disease can continue undetected. A new AI model may help to diagnose these individuals.
Researchers at Kaunas University of Technology (KTU) have developed an AI model that can accurately and objectively analyse a person’s emotional state by monitoring speech and brain neural activity. Traditional depression diagnostics tend to rely on one stream of data, but by using a multimodal approach, the KTU scientists were able to get a better sense of an individual’s emotional state and achieve over 97 per cent in diagnostic accuracy.
The research team chose to use data collected from the voice because a person’s emotional state can be revealed through the pace of speech, intonation, and overall energy. This data was collected during question-and-answer sessions, where the patients also completed activities like describing pictures so the researchers could capture their speech.
In addition to the patient’s voices, the researchers also compiled electroencephalogram (EEG) data from the Multimodal Open Dataset for Mental Disorder Analysis (MODMA). The participants were recorded for five minutes while they were awake, at rest, and with their eyes closed to measure electrical activity in the brain. This EEG and voice data were then transformed into spectograms, and a deep learning model was used to detect signs of depression and split the data into classes of healthy or depressed people.
The AI model requires further clinical trials and improvements to the programme, including having the algorithm provide specific information on why a diagnosis of depression is made, so that medical professionals can act accordingly. The team believes that, in the future, it has the potential to speed up the diagnosis of depression, enabling remote diagnosis and reducing subjective evaluations.
Written By: Jessica Wallis