Observer and filter approaches for the frequency analysis of speech signals

Language disorders can be classified into the three major linguistic levels of lexicon, grammar, and pronunciation. Most patient-oriented sessions at therapists' offices involve an enormous amount of work that is related to the analysis of spoken language. Therefore, it is desired to develop a software-based assistance system allowing a therapist to focus his/ her valuable time on the actual therapy work. For that reason, a joint research project - bringing together the fields of signal processing and speech therapy - has been started recently. It consists of the following aims: (i) automatic transcription and preprocessing of spoken text involving erroneous pronunciation, (ii) automatic classification of pronunciation disorders, (iii) grammatical analysis of freely spoken language. This paper is related to the first project aim by providing an observer and filter-based substitute for the offline frequency analysis that is currently used in many state-of-the-art language recognition systems. Test cases of natural spoken language show the benefits and advantages of the proposed technique in comparison with the widely used offline Fourier transformation for frequency analysis.