Depression has become a common word. Many ordinary people say they have depression when they are in a bad mood. But even for doctors, diagnosis is not an easy task. Machine learning tools developed by researchers at the University of Southern California may make diagnosis easier and more objective.
The tool, called SimSensei, records the patient's voice in an interview and detects if the vowel expression is reduced because it is a common feature in mental and neurological disorders, but it is difficult for humans to detect. Although this method cannot replace human diagnosis, it also provides an objective standard.
Misdiagnosis of depression is a big problem. A review study in 2009 found that only half of the 50,000 patients were correctly diagnosed, and the false positives and false negatives were 3:1. There are two reasons for this: First, misdiagnosis is safer than no diagnosis, and second, it requires more expertise to determine the possibility of any diagnosis.
For depression, it is a heterogeneous disease with multiple causes and different forms of expression. Coupled with the fact that doctors may see hundreds of patients with different conditions and different descriptions in a week, misdiagnosis can be said to be justifiable. That's why tools like SimSensei can do a bigger job.
Previous investigations have found that people with depression are more blunt and negative, voice changes are reduced, volume and monotony are more monotonous, speech is reduced, and speech is unclear, and pauses become longer. In addition, the vocal cords and vocal cords of depressed patients are more intense. Machine learning is well suited to solve such problems, can be predicted from noise data, and speech analysis is also an important topic in this field.
The principle is simple, the patient's speech is processed into only vowels, and then the first and second formants (spectral peaks) of the vowel a/i/u are analyzed. Finally, the k-means algorithm is used for processing. This algorithm is quite old. It appeared in 1967. The principle is to divide the data set into different classes around a certain average value.
The result of the clustering is a triangle, with the corners representing the peaks of the vowels. The region within the triangle represents the vowel space, which is compared to a standard vowel space used for comparison, and the resulting ratio can be used for depression diagnosis.
The effect of SimSensei has also been proven, and the results show that the effect is good when the voice data is limited, which shows that it has certain practicability.
08*16 2835 Led Neon Flex,Neon Flexible,Neon Flex Light,Neon Flex Led Strip
NINGBO SENTU ART AND CRAFT CO.,LTD. , https://www.lightworld-sentu.com