Phoneme classification using wavelet packets

This paper shows that wavelet packets can be used to classify and identify phonemes. The experiment conducted uses the divide and conquer approach. The entire set of voiced and unvoiced phonemes in English language, which accounts to around 44, is grouped into small sets using wavelet packets. Later, these individual groups are examined to determine the exact phoneme. Classification and Identification of phonemes is done by comparing the energy spectrum. Our algorithm is tested on 2 subjects a male and a female originating from different regions, with a difference in their age, sex, microphone used for recording speech, the environment in which these recordings were done etc. Experiment on these subjects resulted in an overall identification accuracy of 82%.