Spiking network optimized for noise robust word recognition approaches human-level performance and predicts auditory system hierarchy
The auditory neural code is resilient to acoustic variability and capable of recognizing sounds amongst competing sound sources, yet, the transformations enabling noise robust abilities are largely unknown. We report that a hierarchical spiking neural network (HSNN) trained to maximize word recognition accuracy in noise and multiple talkers approaches near human-level performance. Intriguingly, comparisons with data from auditory nerve, midbrain, thalamus and cortex reveals that the organization and nonlinear transformations of the optimal network predict several properties of the ascending auditory pathway including a sequential loss of temporal resolution, increasing sparseness and selectivity. The optimal organizational scheme is critical for noise robustness since an identical network arranged to enable high information transfer does not predict auditory pathway organization and has substantially poorer performance. Furthermore, conventional linear and nonlinear receptive field-based models fail to achieve similar noise robust performance. The findings suggest that the auditory pathway hierarchy and its sequential nonlinear feature extraction computations may form a near optimal code capable of efficiently detecting sounds in noise impoverished conditions.