Amy Beeston comes form the group of Guy J. Brown (University of Sheffield, UK) and speaks about "Compensation for reverberation using auditory models".
Abstract:
In contrast to the conventional understanding of reverberation as an unwanted signal distortion, researchers have recently demonstrated that exposure to reverberation can at times improve speech perception.
However, although a growing body of psychoacoustic evidence now replicates this compensation effect, the physiology underpinning our robustness to reverberation so far remains unknown.
A candidate model of perceptual compensation for reverberation is introduced, based on psychoacoustic principles of the human auditory system. Two main assumptions underpin the model design: (i) that the effects of late reverberation can be understood to be somewhat similar to those of additive noise (resulting in an increased noise floor and reduced dynamic range); and (ii) that the medial olivocochlear (MOC) efferent system, known to be active when listening to speech in noise, might also be involved therefore when listening to reverberant speech.
The model embeds an efferent-inspired cochlear filter bank [Ferry and Meddis (2007), J Acoust Soc Am. 122, 3519-3526] within a feedback loop, so that the encoding of sound is regulated in response to an assessment of the level of reverberation observed in the simulated auditory nerve response. Using a simple template-based speech recogniser, the model approximates human compensation data observed in Watkins' monaural phoneme-continuum experiments [Watkins (2005), J Acoust Soc Am. 118, 249-262].
A series of perceptual experiments additionally confirms that monaural compensation occurs with naturalistic speech stimuli [Beeston et al (2014), J Acoust Soc Am. 136, 3072-3084], and addresses questions arising during the modelling process. These include the temporal extent of the ‘context window’ over which the reverberation content should be judged, and the desirable characteristics of a metric which automatically determines MOC suppression. Implications of the psychoacoustic findings on future auditory modelling are explored, and their potential relevance for hearing-impairment and reverberation-robust speech processing is discussed.