Neural correlates of the perception of sound source separation.

As two sound sources become spatially separated in the horizontal plane, the binaural cues used for sound localization become distorted from their values for each sound in isolation. Because firing rates of most neurons in the inferior colliculus (IC) are sensitive to these binaural cues, we hypothesized that these neurons would be sensitive to source separation. We examined changes in the target azimuth tuning functions of IC neurons in unanesthetized rabbits caused by the concurrent presentation of an interferer at a fixed spatial location. Both target and interferer were broadband noise bursts, uncorrelated with each other. Signal detection analysis of firing rates of individual IC neurons shows that responses are correlated with psychophysical performance on segregation of spatially separated sources. The analysis also highlights the role of neural sensitivity to interaural time differences of cochlea-induced envelopes in performing this task. Psychophysical performance on source segregation was also compared to the performance of two contrasting ­maximum-likelihood classifiers operating on the firing rates of the population of IC ­neurons. The "population-pattern" classifier had access to the firing rates of every neuron in the population, while the "two-channel" classifier operated on the summed firing rates from each side of the brain. Unlike the two-channel classifier, the ­population-pattern classifier could segregate the sources accurately, suggesting that some of the information contained in the heterogeneity of azimuth tuning functions across IC neurons is used to segregate sources.