| This work is a modern approach to dysarthria classification and speech characterization using computerized systems and digital speech processing techniques.; Dysarthria is a perturbation of voluntary control over musculature participating in speech production due to impairments of the peripheral or central nervous systems (PCNS). This name is given to a group of neurological diseases among which are: flaccid dysarthria, spastic dysarthria, ataxic dysarthria, Parkinson's disease, chorea, dystonia, organic voice tremor and amyotrophic lateral sclerosis.; Clinical studies of dysarthria have revealed that there is a cause-effect relationship between the location of the damage in the PONS and the resulting type of dysarthria. The location of the PCNS that is affected by the disease determines the signs and symptoms of neurological impairment and its speech characteristics. Therefore, a correspondence between both can be established.; The use of perceptual judgment to describe the speech perturbations and perform the classification is widely used. This method is sometimes imprecise and very dependent on the experience of clinicians, and therefore, it requires the use of alternative tools to help and improve its performance. The treatment of dysarthria varies depending on the type, thus making the assessment process a crucial step in order to detect and control the evolution of the disease. Consequently, the use of speech processing techniques to support the diagnosis may increase the precision of speech characterization and provides more objective measures for classification.; This research describes a methodology to characterize perturbations of dysarthric speech and develops a classifier based on semiautomatic characterization of speech deviations using recorded speech. The methodology is based on a novel approach to describe dysarthric speech where those features less reliably perceived by clinicians are estimated using digital signal processing techniques, while others are obtained from clinicians and medical records. The classifier is based on self-organizing map; neural networks that perform non-linear analysis on the resulting set of features to identify the dysarthric groups and provide clinicians with reliable clues about the damage in the nervous system. New algorithms for speech characterization were also devised and combined with the novel classification technique to produce an expert system that provides objective information about speech deviations and improve the accuracy of classification of these types of diseases. |
