Differential diagnosis of subtypes of dysarthria is important because it enables clinicians to determine the location of the lesion responsible for the disorder and allows clinician and patient to set appropriate treatment goals. Even more important is a reliable, objective method of documenting treatment efficacy. Perceptual methods of evaluating dysarthria are most common. Yet differences between experts are larger than the differences needed for diagnosis or monitoring and are hard to quantify. We propose an automatic acoustic-feature recognition system to aid clinicians in evaluating dysarthria. The system and two phoneticians will be trained to judge acoustic features from sustained vowels and diadochokinetic syllables spoke by subjects with a variety of dysarthrias. The system will make its judgement from acoustic features like periodicity of glottal pulses or rate and regularity of diadochokinetic syllables, and others. The phoneticians will make their judgements based on perceptual features like breathiness, diplophonia, harshness, harshness, and others. The goal of Phase I is to show that the proposed system can identify the acoustic characteristics of different subtypes of dysarthria at 90 percent accuracy with respect to human judgements of the associated perceptual features.
There is a clear need for an efficient, reliable, and objective tool to evaluate the speech of dysarthric individuals. The software could be packaged with existing acoustic analysis systems which are currently purchased by hospitals, clinics, and universities for non-automatic analysis of dysarthric speech.
