In this paper a method of classifying phonemes by combining a dynamical systems approach with subband decomposition of speech signals is presented. The ability of reconstructed phase spaces to effectively model sub-bands of phonemes in different phonological classes is demonstrated. Experiments performed over the TIMIT database show how well phonemes from different phonological classes can be recognized in different frequency bands. It is hypothesized that given these results, filtering signals before embedding has the potential to improve classification accuracy.
Bibliographic reference. Indrebo, Kevin M. / Povinelli, Richard J. / Johnson, Michael T. (2003): "A combined sub-band and reconstructed phase space approach to phoneme classification", In NOLISP-2003, paper 017.