Speech recognition based on fundamental functional principles of the brain

This paper describes application of biologically realistic dynamic synapse neural networks on speech recognition task. The goal is to develop a noise-robust and feasible speech recognition system, based on fundamental functional principles of the brain. This task is accomplished in three steps: first speech signal is decomposed into different frequency bands. Second, short term energy of signals is encoded into the train of spikes and finally, the classification of temporal patterns is done using dynamic synapse neural networks (DSNN). The nonlinear neurotransmitter release function of DSNN is replaced by FD model. The simulation results showed that the performance degradation of DSNN in the presence of Gaussian white noise is less than Mel frequency cepstral coefficients