Het diagnosticeren van dysartrische spraak door middel van confusion matrices.

Keywords
Loading...
Thumbnail Image
Issue Date
2021-08-30
Language
nl
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
The intelligibility of dysarthric speech is one of the most researched areas. The intelligibility is often measured in different ways. Multiple researchers (Xue et al., 2020) have mentioned that objective measures of speech intelligibility are needed. Confusion matrices have often been used in the past to give an overview of articulation errors such as substitutions, insertions and deletions. The main research question of the current master thesis is: ‘In which way can the confusion matrices, computed from two experiments from previous research by Xue et al. (2020), be used to obtain information about patterns in deviations in dysarthric and healthy speech?’ The confusion matrices have been analyzed both descriptively and statistically. The descriptive and statistical analysis concentrated on the differences between factors such as ‘type of group’ (dysarthric and normal speakers), ‘type of stimuli’ (meaningful words with and pseudowords without contextual information (experiment 1 or 2)) and ‘type of transcription task’ (meaningful (WTrans) vs. meaningful-/pseudowords (LTrans)). Both descriptive and statistical analysis looked at various types of confusion matrices like normalized, unnormalized, matrices of vowels/diphthongs and matrices of consonants. All the confusion matrices were based on graphemes. The highlights of these analysis are described in this thesis report. Different statistical analysis were carried out in the program SPSS (IBM SPSS, version 25). In order to investigate the effect of contextual information (meaningful words with contextual information (experiment 1) vs. meaningful words and pseudowords with no contextual information (experiment 2)) on the number of deviations made and to investigate the effect of transcriptions task (meaningful words (WTrans) v. meaningful-/pseudowords (LTrans) on the number of deviations made, a descriptive analysis and two paired t-tests were computed. To investigate the effect of the severity of the dysarthric speech and type of grapheme on the number of deviations made, two 4 x 5 Mixed ANOVA were computed, both of experiment one and two. It was expected that there would be differences between the mentioned factors. Also, it was expected that there would be a significant effect of ‘type of group’ (dysarthric and normal speakers), ‘type of grapheme’ and interaction-effect ‘type group * type of grapheme’ on the number of deviations. The results of both the descriptive and statistical analysis showed that differences were found between the ‘type of speech’, ‘type of stimuli’ and ‘type of transcription task’. The results of experiment one showed a significant effect of the within-subject factor ‘type of grapheme’ on the number of deviations shown by the relevant graphemes. The results of experiment two showed a significant effect of the factor ‘group’ (dysarthric and normal speakers) and ‘type of grapheme’ on the number of deviations made. In short, these results show that the different confusion matrices can be used to obtain important information of the dysarthric speech. Future research should focus on using confusion matrices on phoneme level instead of grapheme level. Then more can be said about the speech intelligibility of dysarthric speakers. Based on the results on grapheme level, only assumptions can be made about the speech intelligibility of these speakers. Results on phoneme level will give more information about the speech intelligibility and is therefore needed.
Description
Citation
Faculty
Faculteit der Letteren