Abstract:

Visual images have considerable value in freezing rapidly changing speech signals, both for the analysis of disputed speech samples and in the presentation of such material in court. Speech spectrograms are routinely used in speech analysis to examine the speech signal as a whole, prior to a more systematic and directed examination of particular speech characteristics, the forensic significance of which requires forensic phonetic expertise. This paper presents spectrograms based on peripheral human hearing modelling to illustrate their potential usefulness for speech analysis and discusses current developments towards a real-time version of such a system. The main differences between these and spectrograms derived by traditional means are discussed.

Author: David M. Howard, Allen Hirson, Tim Brookes, Andrew M. Tyrrell