Binding of audio elements in the sound source segregation problem via a two-layered bio-inspired neural network

We use a two-layered bio-inspired neural network to segregate sound sources, i.e. double-vowels or intruding noises in speech. The architecture of the network consists of spiking neurons. The spiking neurons in both layers are modelized by relaxation oscillators. The first layer of the network is locally connected, while the second layer is a fully connected network. Our auditory image is based on the reassigned spectrum technique. No prior estimation or knowledge of pitch is necessary for the segregation.