Artificial Spectro-Temporal Receptive Fields Evoked by Long Speech Signals

Neurons in the primary auditory cortex A1 unitize multidimensional receptive activation fields, which respond to specific frequency bands and timing patterns of the input signal. Such ‚spectro-temporal receptive fields’ (STRF) can experimentally be determined by reverse correlation methods if the activation or output signal patterns of the neurons are known. This seems impossible for neurons in the human auditory cortex; Thus, only simulations can be done, which show that an evolution of respective banks of STRF is possible or convergent and useful from the perspective of ‚optimal’ information processing. In [1], an STRF simulation by Gabor functions or sigma-pi artificial neurons was used to find robust cues for traditional sound processing algorithms. An alternative STRF-based sound processing scheme can be found in [2].