Performance analysis of novel speech based psychological assessment tool using Bahasa Malaysia /
Major depressive disorder is a global growing cause for concern. In pursuit of reducing the statistics, since the past few decades, researchers have been working on producing automatic objective screening mechanism using biometric parameters. One of the possible parameters in diagnosing psychologica...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
Kuala Lumpur :
Kulliyyah of Engineering, International Islamic University Malaysia,
2017
|
| Subjects: | |
| Online Access: | Click here to view 1st 24 pages of the thesis. Members can view fulltext at the specified PCs in the library. |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Major depressive disorder is a global growing cause for concern. In pursuit of reducing the statistics, since the past few decades, researchers have been working on producing automatic objective screening mechanism using biometric parameters. One of the possible parameters in diagnosing psychological state is speech, which is dependent on many factors such as language and speakers. However, to date, none of the research was done using speech characteristics in Bahasa Malaysia native speakers. This research hereby sought to identify possible acoustic features that can be used as an indicator for depression using speech in Bahasa Malaysia. Since the characteristics of male and female speech are different, the data was analysed separately. We obtained clinically validated data of six depressed and ten healthy subjects for male, and seven depressed and ten healthy subjects for female. Four types of acoustic features were extracted, namely Mel Frequency Cepstral Coefficient (MFCC), Power Spectral Density (PSD), Transition Parameters and Interval length Probability Density Function (PDF). Linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA) were used to obtain the decision boundary for the pairwise classification with resampling techniques of jack-knife and cross validation. Using only a single feature to train the classifier model, we found that the first cepstral coefficient, MFCC-C1 outperformed the rest with 93.2% in accuracy when classified using LDA for male data and 95% in accuracy when classified using both LDA and QDA for female data. Remarkably, it can be concluded that information contained in MFCC-C1 is robust across gender. To get the optimum feature combination, the features were then combined with the maximum number of three. For both datasets, 100% accuracies were achieved by classifiers trained using combinations of two features. Considering the difficulty in acquiring depression speech database, more effort shall be put into collecting more data and validating the analysis. This work demonstrates an optimistic implementation of the desired objective diagnostic tool as it proves that there are distinctive patterns in depressed and healthy datasets of Bahasa Malaysia speakers. |
|---|---|
| Physical Description: | xvi, 89 leaves : colour illustrations ; 30cm. |
| Bibliography: | Includes bibliographical references (leaves 72-77). |