Efficient coding of natural sounds

Introduction Many previous studies in auditory neurophysiology have used simple tonal stimuli to understand how neurons encode sound. While these studies have shed light on many important neural characteristics, pure tones do not typically occur in our environment. Instead, we are often surrounded by multiple sound sources with complex harmonic and transient components such as speech, environmental sounds, animal vocalizations, and background noise. These natural sounds only make up a small subset of the sample space of all possible acoustic stimuli, yet they still consist of a wide range of spectral and temporal structures. It is reasonable to hypothesize that our brains have evolved to optimally process these naturally occurring sounds in order to efficiently extract relevant acoustic cues. In recent years, an increasing number of modeling and physiological studies have used natural or natural-like stimuli to explore the validity of this prediction. With this better understanding of natural sound statistics and methods of decomposing their signals, future studies can create synthetic stimuli with similar statistics to investigate (and possibly differentiate) between natural and “naturalistic” sounds. This brief overview of the coding of natural sounds suggests three papers for discussion. The first demonstrates that the statistics of natural sounds are redundant in the peripheral auditory system representation (Attias and Schreiner 1997). The remaining two studies implement two different ways of decomposing a sound signal. One study shows that a Fourier analysis may be sufficient for animal vocalizations, while wavelet transforms are optimal for encoding speech and environmental sounds (Lewicki 2002a). The other uses modulation spectra to encode natural stimuli and demonstrates differences between groups of natural sound ensembles (Singh and Theunissen 2003). It is also useful to interpret the findings of these studies in terms of neural responses to natural sounds. Two physiological studies in zebra finches and grasshoppers are also proposed for further reading (Hsu et al 2004, Machens et al 2005).