Title :
Is speech chaotic?: invariant geometrical measures for speech data
Author :
Banbrook, Michael ; Mclaughlin, Steve
Author_Institution :
Signals & Syst. Group, Edinburgh Univ., UK
Abstract :
A set of prolonged vowel sounds have been analysed to reveal if there are any underlying low dimensional dynamics, and if so quantify them. The approach used was to embed the time series data into a d-dimensional state space using time delay embedding with a reconstruction delay given by mutual information analysis. The correlation dimension has been shown to vary between one and three across the data set and the results suggest that there may be a connection between the correlation dimension and the manner of vowel articulation. The Lyapunov exponents have been calculated and although the results are not conclusive they do point towards the existence of positive exponents suggesting that the system is chaotic and placing limits on the predictability of speech signals
Keywords :
Lyapunov methods; chaos; speech processing; state-space methods; time series; Lyapunov exponents; chaos; correlation dimension; invariant geometrical measures; low dimensional dynamics; prolonged vowel sounds; speech signals; state space; time delay; time series data;
Conference_Titel :
Exploiting Chaos in Signal Processing, IEE Colloquium on
Conference_Location :
London