Abstract:

Previous research in speaker characteristics has tended to focus on static properties of the speech signal. Static features of speech generally display differences between individuals, but are insufficient to characterize a speaker. The present study considers reasons why dynamic features of speech should provide an important source of speaker-distinguishing information. It discusses this idea for formant frequencies in particular, and outlines a new direction for the development of a technique for characterizing individual speakers which uses regression to parameterize formant frequency contours. The technique is trialled and results reported for two data sets.

Author: Kirsty McDougall