Scibelli, Filomena (2018) Detection of Verbal and Nonverbal speech features as markers of Depression: results of manual analysis and automatic classification. [Tesi di dottorato]

[img]
Preview
Text
thesisPhdFilomenaScibelli_sub2.pdf

Download (2MB) | Preview
[error in script] [error in script]
Item Type: Tesi di dottorato
Resource language: English
Title: Detection of Verbal and Nonverbal speech features as markers of Depression: results of manual analysis and automatic classification
Creators:
CreatorsEmail
Scibelli, Filomenafilomena.scibelli@libero.it
Date: 31 October 2018
Number of Pages: 134
Institution: Università degli Studi di Napoli Federico II
Department: Studi Umanistici
Dottorato: Mind, Gender and Language
Ciclo di dottorato: 31
Coordinatore del Corso di dottorato:
nomeemail
Bacchini, Dariodario.bacchini@unina.it
Tutor:
nomeemail
Esposito, AnnaUNSPECIFIED
Vinciarelli, AlessandroUNSPECIFIED
Date: 31 October 2018
Number of Pages: 134
Keywords: Depression; speech prosody analysis; verbal content analysis; automatic classification; Support Vector Machine
Settori scientifico-disciplinari del MIUR: Area 11 - Scienze storiche, filosofiche, pedagogiche e psicologiche > M-PSI/01 - Psicologia generale
Date Deposited: 28 Dec 2018 17:45
Last Modified: 30 Jun 2020 09:08
URI: http://www.fedoa.unina.it/id/eprint/12462

Collection description

The present PhD project was the result of a multidisciplinary work involving psychiatrists, computing scientists, social signal processing experts and psychology students with the aim to analyse verbal and nonverbal behaviour in patients affected by Depression. Collaborations with several Clinical Health Centers were established for the collection of a group of patients suffering from depressive disorders. Moreover, a group of healthy controls was collected as well. A collaboration with the School of Computing Science of Glasgow University was established with the aim to analysed the collected data. Depression was selected for this study because is one of the most common mental disorder in the world (World Health Organization, 2017) associated with half of all suicides (Lecrubier, 2000). It requires prolonged and expensive medical treatments resulting into a significant burden for both patients and society (Olesen et al., 2012). The use of objective and reliable measurements of depressive symptoms can support the clinicians during the diagnosis reducing the risk of subjective biases and disorder misclassification (see discussion in Chapter 1) and doing the diagnosis in a quick and non-invasive way. Given this, the present PhD project proposes the investigation of verbal (i.e. speech content) and nonverbal (i.e. paralingiuistic features) behaviour in depressed patients to find several speech parameters that can be objective markers of depressive symptoms. The verbal and nonverbal behaviour are investigated through two kind of speech tasks: reading and spontaneous speech. Both manual features extraction and automatic classification approaches are used for this purpose. Differences between acute and remitted patients for prosodic and verbal features have been investigated as well. In addition, unlike other literature studies, in this project differences between subjects with and without Early Maladaptive Schema (EMS: Young et al., 2003) independently from the depressive symptoms, have been investigated with respect to both verbal and nonverbal behaviour. The proposed analysis shows that patients differ from healthy subjects for several verbal and nonverbal features. Moreover, using both reading and spontaneous speech, it is possible to automatically detect Depression with a good accuracy level (from 68 to 76%). These results demonstrate that the investigation of speech features can be a useful instrument, in addition to the current self-reports and clinical interviews, for helping the diagnosis of depressive disorders. Contrary to what was expected, patients in acute and remitted phase do not report differences regarding the nonverbal features and only few differences emerges for the verbal behaviour. At the same way, the automatic classification using paralinguistic features does not work well for the discrimination of subjects with and without EMS and only few differences between them have been found for the verbal behaviour. Possible explanations and limitations of these results will be discussed.

Downloads

Downloads per month over past year

Actions (login required)

View Item View Item