Diagnosis of dysarthria subtype via spectral and waveform analysis

This study reports on the performance of a computerised digital signal processing system, known as the Computerised Frenchay Dysarthria Assessment (CFDA), which is designed to diagnose two sub-types of dysarthria – a family of speech disorders characterised by loss of control over the organs which facilitate speech production. This investigation explores the use of both spectral and waveform analysis to distinguish between the ataxic and mixed dysarthria subtypes by assessing articulatory competence in the execution of two speech-related tasks. It is demonstrated that waveform analysis of utterances representing the consonant /p/ can reliably measure a speaker’s lip seal competence; this combination of lip seal evaluation and voice quality measurement can then serve as a composite tool to detect certain pathological features characteristic of ataxic and mixed dysarthria respectively. To validate this hypothesis, this study compares the assessment accuracy of the CFDA application with that of a panel of expert clinicians when evaluating a series of speech samples from a selection of individuals, some of whom were previously diagnosed as suffering from either ataxic or mixed dysarthria. The CFDA’s diagnostic output from this data evaluation exercise produced an overall correlation of 0.91 with those of the expert clinicians. This close correlation reinforces the validity of the objective voice quality evaluation procedures developed during the course of this study.