Processing and classification of deformed speech using neural networks

In many problems of medical diagnosis as well as therapy and rehabilitation, evaluation of the deformed speech signal quality is required. In problems of distorted speech diagnosis the regular methods of speech signal preprocessing and classification, used in speech or voice recognition, totally fail. Also the standard speech signal parametrization techniques (e.g. LPC or cepstral coefficients) cannot satisfactorily describe pathological speech because of its dissimilar phonetic and acoustic structure compared with correct speech, and also because the aim of the recognition process is totally different. In the paper a new method for processing and classification of pathologically deformed speech, based on neural networks techniques, is presented and discussed.