Perceptual Compensation for Reverberation: Effects of ‘Noise-Like’ and ‘Tonal’ Contexts

Abstract

Listeners’ perceptual decisions about the characteristics of speech sounds that are heard in rooms appear to utilize information about the room’s acoustic properties. This is shown in experiments on perceptual compensation for room reverberation, where the mechanism responsible seems to pick up information about reverberation from a running-speech ‘context’, and uses it to ameliorate effects of the reverberation on neighbouring test words (Watkins 1992, 2005). Some of these compensation experiments have used the [s] vs [st] distinction between “sir” and “stir” test-words and the context phrase “next you’ll get _ to click on”. When reverberation in the context is kept at a minimal level, reverberation added to test words generally makes them sound more like “sir”. This effect arises because of the smoothly-decreasing energy-decay that is typical of reverberation in box-shaped rooms (Schroeder 1965; Allen and Berkely 1979). Such reverberation adds a ‘tail’ to the [s], which obscures the gap that is a major cue to the presence of a [t]. If this happens, compensation may be seen when the amount of reverberation in the context is also increased, to a level close to the amount in the test word. The effect of this compensation is to make the test words sound more like “stir” again. The present experiments ask about the characteristics of context sounds that are important in effecting compensation. Sharp offsets in a sound are a likely source of information about reverberation (Stecker and Hafter 2000) because they will be prominent if the sound is relatively ‘dry’, or extended to form smooth ‘tails’ when there is more reverberation. Sharp offsets can occur at the ends of sounds, but they can also occur in narrow frequency-bands during continuous speech if the band’s power falls abruptly due to a change in the shortterm spectrum (‘spectral transition’, Furui 1986). Narrow-band offsets of this type are therefore likely to be prominent in continuous speech, as the inherently dynamic origins of this signal give rise to numerous spectral transitions. Such offsets might therefore provide listeners with information about a room’s