“Clinical depression is, simply put, a dreadful disease. Diagnosing it is anything but simple, however. Its symptoms vary, can shift with the ups and downs of everyday life, and sometimes overlap with those of other diseases. For these reasons, it is common for depression to go unidentified for months, or even to be missed altogether.
Stefan Scherer of the University of Southern California and Louis-Philippe Morency of Carnegie Mellon University, in Pittsburgh, hope to change this. They are trying to develop a reliable way of diagnosing depression by using a computer to record and analyse aspects of a putative sufferer’s behaviour. They are, they think, 85% of the way there.
Their latest research, just published in IEEE Transactions on Affective Computing, is an analysis of depressed people’s speech. It follows up on analyses of facial expressions and eye movements carried out by tracking cameras while the subject of a diagnosis is having a conversation. These used things like the length (or, rather, shortness) of people’s smiles and the frequency with which they looked at the ground in order to develop an algorithm that was 75% effective in diagnosing depression.” said economist.com
“The extra 10% of reliability has come from quantifying what was previously a qualitative observation, which is that depressed people tend to run their vowels together when they speak. Dr Scherer and Dr Morency programmed their software to record patterns of vowel-spacing (known as vowel-space ratios) and then tested the system on more than 250 people, some of whom had been diagnosed independently as depressed and some of whom had not.
The software found that depressed participants’ vowel-space ratio in normal speech was 0.49 (compared with 1.0 for a reference population reading a standard word list). That of healthy participants averaged 0.55. A small difference, certainly—and not enough by itself to be a reliable diagnostic tool. But combined with the researchers’ previous work, this was a palpable advance.” said economist.com