| Natural sounds, such as speech and vocalizations, are characterized by time-varying spectra that give rise to distinct temporal periodicities, frequency transitions, and spectral resonances. These structural features are decomposed by the primary sensory epithelium and give rise to a number of perceptual attributes. Given the complexity of the auditory neuronal network and the fact that the brain is in general extremely nonlinear, it is increasingly clear that simple acoustic stimuli (e.g. pure tones and noise) can not be used to identify natural sounds processing strategies. To understand how complex sound attributes are represented in the brain, statistical properties of the spectro-temporal envelope of natural sounds (including speech, vocalizations, environmental noise, and music) were studied in detail. Ensemble statistics are marked by robust spectrographic correlations, logarithmic contrast, and stimulus dynamics which are closely related to a number of perceptually relevant acoustic variables. Hypothetically, these higher-order stimulus attributes can be utilized by the auditory system for efficient sound processing and across-category discrimination. To test this hypothesis, neuronal recordings were performed in the central nucleus of the inferior colliculus (ICC) of cats using synthetic ripple stimuli that incorporate the observed statistical attributes. Using spectro-temporal receptive field (STRF) methods, it is found that ICC neurons efficiently utilize these higher-order stimulus attributes for sound processing. Populations of neurons are distinguished based on their degree of feature selectivity and their ability to time-lock to the spectro-temporal envelope. A hierarchy of functionally distinct neuronal types is revealed based on three possible neuronal codes. Further evaluation reveals that the operating range of ICC neurons is physically matched to the spectro-temporal energy distributions observed in natural sounds. When tested with stimuli that mimic natural sounds, neurons show contrast tuning and improved spectro-temporal coding at time-scales comparable to the neuron's receptive field. These findings establish a link between acoustic ecology, acoustic sound structure, and neuronal processing. Such processing strategies make use of structural regularities in natural sounds and likely underlie human perceptual abilities. |
