Exploitation of Phase Information and Non-Spectral Features in Signal Mining

Frame-based analysis of signals is often conducted based on the magnitude

spectrum only and computed by using such frame lengths that signal details

are lost. With the aim of estimating additional information during signal

analysis in general, this research investigates signal phase derived

information, e.g. average phase slope and group delay, by identifying and

utilising epochs in a signal that are important for tracking natural

variations in the underlying source. The work will facilitate detailed

signal analysis on basis of epoch synchronised frames of small and/or

dynamic length and can potentially be applied on a variety of signals,

e.g. biomedical and speech. The research leads to multi-disciplinary

efforts among medical doctors, phoneticians and engineers within dedicated

application areas.