Overlapping ‘what’ and ‘where’ in a model of auditory cortical processing

.The cues for identifying and localizing auditory events overlap in the frequency spectra of sounds resulting in a demanding computational challenge. To study how these cues might be extracted by the auditory system, we trained a neural network model to localize speech sounds. The resulting network utilized neurons sensitive to speech sound identity to localize sound sources. Our results suggest that the strategy used by the auditory cortex to perform ‘where’ tasks may also involve extracting ‘what’ information.