Study of Attractor Variation in the Reconstructed Phase Space of Speech Signals

This paper presents a study of the attractor variation in the reconstructed phase spaces of isolated phonemes. The approach is based on recent work in time-domain signal classification using dynamical signal models, whereby a statistical distribution model is obtained from the phase space and used for maximum likelihood classification. Two sets of experiments are presented in this paper. The first uses a variable time lag phase space to examine the effect of fundamental frequency on attractor patterns. The second focuses on speaker variability through an investigation of speaker-dependent phoneme classification across speaker sets of increasing size.