PhD Scientific Days 2020

Mental Sciences I. Posters

Examination of Acoustic Features in Depression, Developing an Automatic Decision System for Discriminating Speech Pathology

Előadó neve

Dr. Hajduska-Dér, Bálint

Előadó munkahelye

Semmelweis Egyetem Pszichiátriai és Pszichoterápiás Klinika

Előadó telefonszáma

+36208250224

Előadó e-mail címe

hajduska-der.balint@med.semmelweis-univ.hu

Az előadás címe

Examination of Acoustic Features in Depression, Developing an Automatic Decision System for Discriminating Speech Pathology

Szerző(k) neve és munkahelye

1 Bálint Hajduska-Dér, Department of Psychiatry and Psychotherapy, Semmelweis University, Budapest
2 Gábor Kiss, Department of Telecommunications and Media Informatics, University of Technology and Economics, Budapest
3 Klára Vicsi, Department of Telecommunications and Media Informatics, University of Technology and Economics, Budapest
4 Lajos Simon, Department of Psychiatry and Psychotherapy, Semmelweis University, Budapest

Szekció

Mental Sciences I. Posters

Language of the presentation

Hungarian

Section, first choice

Mental Sciences

Section, second choice

Clinical Medicine

Összefoglaló szövege

Introduction: WHO studies show that the prevalence of major depression increased 18,4% in the last ten years. The early diagnose and adequate treatment of depression decrease suicidal risk and mortality, which shows a great need for new and fast diagnostical methods. The effects of depression in speech are studied with computer sciences in the past decades. Before that era physicians could only tell about the qualitative features of speech like speed, intensity, pauses. In the past few years, machine learning was introduced in diagnostical studies, with the ability to find regularities in big data. In our study, we developed a support vector regression-based machine learning system for discriminating the speech of patients with depression.
Aim: Our purpose is to develop an automatic decision system, that can be used to separate speech samples of patients with depression from healthy speakers and give a probability of the severity of depression.
Method: Speech samples were collected from patients diagnosed with depression. The severity of depression was assessed by Beck Depression Inventory-II (BDI) and Hamilton Depression Scale (HAM-D). Patients on antipsychotic medication were left out of the study, because it’s probable effect on the acoustical features of speech. The samples contained a read text called “The North Wind and the Sun”, each speech was segmented on phonema level and support vector regression was used in the automatic decision system.
Results: In this phase of the study we introduced the HAM-D to objectify the diagnose of depression and improve the accuracy of the automatic decision system, which was previously built with 158 speech samples of patients with depression. With the use of HAM-D, in a current database of 22 patients, we could increase the accuracy (91%) and sensitivity (95%) in speech analysis.
Conclusion: The results of our study show that the acoustic biomarkers of depression can be a viable diagnostical tool in the early recognition of depression. With machine learning, an automatic decision system in speech analysis can be helpful in general medical practice as a screening process for depression. In psychiatry, this system can speed up the initiation of proper treatment and can be used as an objective indicator to measure the effectiveness of variable form of therapies and keep track of the change during treatment.

Additional Information

Supervisor: Lajos Simon
E-mail address: simon.lajos@med.semmelweis-univ.hu

Bemutatás módja

Poszter

Kind

Szabad

Status

elfogadva

Accepted presentation method

poszter

Előadás fájl jóváhagyás

nem rendelkezett róla

Előadó

4669

Start

11:22

End

11:25

Authors (legacy)

1 Bálint Hajduska-Dér, Department of Psychiatry and Psychotherapy, Semmelweis University, Budapest
2 Gábor Kiss, Department of Telecommunications and Media Informatics, University of Technology and Economics, Budapest
3 Klára Vicsi, Department of Telecommunications and Media Informatics, University of Technology and Economics, Budapest
4 Lajos Simon, Department of Psychiatry and Psychotherapy, Semmelweis University, Budapest