Classification in the Speech Recognition
This paper presents specific approach of the classification of the incorrect speech sounds. Sampled data are vowels gathered during the speech therapy with children that have difficulties to pronounce them correctly. Continuous wavelet transformation has been applied on these incorrectly pronounce vowels using Morlet wavelet. Coefficients have been analyzed in the context of three main formants that characterized each of the vowels. The selected coefficients have been classified into main clusters, and have been compared with the one obtained for correct signals. At the end some improvements have been proposed in order to use results in the daily speech therapy and to automate process. Data that is used in analysis is gathered during the speech therapy with the children of age between 12 and 14 years. Session is supervised by speech therapist that listens and guides patient to pronounce vowels correctly. The process is very time consuming and will benefit from some elements of automation and provision of feedback to patients by the expert system. Sound features of language can be viewed in several ways. The first way is that the signals are grouped in terms of distribution of acoustic energy in the resonant field, which is caused by speech, which is the acoustic side of the problem, and this approach is given in this paper. Another way is to look at the voice signals from the point of hearing and the ability to receive those signals and to implement parts of the brain, in