NONLINEAR STATISTICS OF HUMAN SPEECH DATA

Authors:

Abstract
We study the structure of point clouds obtained as time delay embeddings of human speech signals by approximating the data sets with certain simplicial complexes and analyzing their persistent homology. Results for several different sounds are presented in embedding dimensions 3 and 4. The first Betti number allows a coarse classification of sounds into three groups: vowels, nasals and noise.

This publication has 5 references indexed in Scilit: