Comparison of automatic detection of depression based on spontaneous and read speech

OData support
Supervisor:
Dr. Vicsi Klára
Department of Telecommunications and Media Informatics

Abstract

The depression is one of the most frequent mental disorders that affects about more than hundred million people worldwide. The most difficult thing for the doctors is to recognize this disease. Because of these facts it is important to find a method wherewith could be much easier to make a diagnosis.

I joined to an ongoing research of BME-TMIT. The topic of my Thesis is the comparison of automatic detection of depression based on spontaneous and read speech. The depressed speech was recorded in Department of Psychiatry and Psychotherapy. First, I had to segment the spontaneous and read speech, too. Then preprocess the voice samples. After the preprocess, I determined the F0, formant frequencies, jitter, shimmer, intensity, speech rate, articulation rate, relative pause length and mel-filters values. I analyzed these parameter’s values between healthy and depressed speech with independent-samples t-test, separately the spontaneous and the read speech. Then I choosed those parameters, which was significant and I examined the selected variables with paired sample t-test, because of I can determine to have a significant difference between spontaneous and read speech, analyzed separately the healthy and the depressed samples. Finally, I created classification with significant parameters of independent-samples t-test with SVM.

My project includes important scientific literature of the subject, definition of the acoustic parameters which I used to the analysis, the way of the preprocess of voice files, the results of the investigation, and the conclusion.

Downloads

Please sign in to download the files of this thesis.