TNO Human Factors Res. Inst., Kampweg 5, 3769 De Soesterberg, The Netherlands
The starting point is the representation of speech in the spectro-temporal domain. From this representation, a number of interesting physical characteristics can be derived on the dynamic properties of speech. These include the temporal and spectral envelopes, correlation patterns, and synchrony. The perceptual relevance of these characteristics will be illustrated by reviewing a number of psychophysical data from experiments including speech-like sounds or specific laboratory-made dynamic stimuli. Within the context of low bit-rate coding and automatic recognition, the ultimate goal of this comparison between speech characteristics and psychophysics is to help identify the cues in the speech signal which are perceptually significant.