Perceptual constancy in real-room listening by humans and machines

Our perceptual 'systems' allow us reliably to judge properties of things in the real world under diverse conditions, as they exhibit 'constancy'. For example, a white surface in dim light can be distinguished from a black surface in bright light, even though the luminance of the two surfaces might be the same. Although constancy is clearly vital for survival, and has been extensively studied in vision, it has not been investigated in hearing very much. This lack of knowledge probably accounts for the poor performance of the current generation of artificial listening devices, which are becoming increasingly important in hearing aids, as well as in other applications of automated speech recognition. We aim to measure the different listening conditions effected by real rooms and then to investigate constancy in hearing with perceptual experiments. This information will then be incorporated into prototype artificial-listening devices, which will be tested for their effectiveness in dealing with the real world conditions that human hearing seems to cope with so exquisitely.