A bio-inspired sound source separation technique in combination with an enhanced FIR gammatone analysis/synthesis filterbank

A sound source separation technique based on a two-layered bio-inspired spiking neural network and an enhanced gammatone analysis/synthesis filterbank is proposed. One of the two bio-inspired proposed spectral maps (Cochleotopic / AMtopic or Cochleotopic / Spectrotopic) is used as a front-end to the neural network depending on the nature of the intruding sound. We show that the use of an FIR gammatone filterbank outperforms the previous results obtained by using an IIR gammatone cochlear filterbank, since the FIR implementation has near-perfect reconstruction ability and the cascade of the analysis and synthesis filterbanks is linear-phase.