An Analog VLSI Integrate-and-Fire Neural Network for Sound Segmentation.
Mark Glover, Alister Hamilton, Leslie S. Smith · 1998
This paper presents a cascadable aVLSI integrateand -fire neural network chip (SPIKE I) capable of realistic biological time constants incorporated into a real time software based sound segmentation system with results. The sound segmentation system is based on an engineering abstraction of the functionality of the cochlea and auditory nerve. A comparison of the software simulation and software /hardware combination results indicates that clustering does occur. Furthermore the patterns of onsets and offsets generated are broadly similar. Analysis of the results indicates area's for improvement. These have been included in a second integrate-and-fire neural network chip (SPIKE II) presently being fabricated. 1 Introduction This paper provides an overview of the sound segmentation system, the integrate-and-fire neural model used and the architecture of the neural network implemented. A comparison using software and hardware implementations with real time data is performed. This suggests...