Presenter

Mr Tomas Dekens - VUB-ETRO [Email]

Abstract

Dysarthria refers to speech disorders that are characterized by a reduced control of the muscles that play a role in the production of speech. This impaired control leads to speech that can be hard to understand. Therapy is needed to improve the intelligibility of dysarthric patients and as such improve their quality of life.
This dissertation deals with computer-assisted therapy of dysarthria in terms of prosody. More specifically, it focusses on the development of algorithms that objectively analyze the prosody of an utterance. These algorithms can be integrated in a therapy tool that records speech and gives feedback about the appropriateness of the prosody of the utterance. Such a tool can be very useful for therapists. In addition, it can allow patients to train their prosodic skills at home.
Statistical analyses were executed to investigate the influence of different prosodic functions on features that characterize a speech waveform. Support Vector Machines (SVM) were trained using features that describe the relevant properties and use human perception as the ground truth for the classification. These SVMs can be used as classifiers that provide direct feedback about prosody.
Some of the proposed algorithms have been integrated in a therapeutic software tool, together with exercise material that has been selected for the therapy of dysarthria. This tool is called Prosodietrainer and an overview of the functionality of this tool is given in this dissertation.
In order to improve the noise robustness of speech applications, a technique that fuses a body-conducted speech signal with a conventional signal is proposed. Also the influence of noise on speech activity detection is investigated. While these algorithms have not been used in combination with dysarthric speech, they certainly possess the potential to allow the prosodic analysis algorithms to be used in noisy environments by adding them as a front-end to the overall system.

Short CV

Civil Electrotechnical Engineer, VUB