Phoneme analysis based on quantitative and qualitative entropy measurement

In this paper two types of redundancy analysis, one based on linear redundancy and the other based on nonlinear entropy-based redundancy, are applied to the OTAGO speech corpus in order to classify the phonemes according to their nonlinearity properties. The redundancy curves are compared both in qualitative as well as in quantitative manner, the latter way involving the generation of two kinds of surrogate data. The results indicate the existence of trivial and dynamic nonlinearities in certain phonemes.