An Effective Method to Decrease the Dimension of Input Vector of BPNN on ASR System

This paper uses BP neural network for modeling and recognition in the ASR (automatic speech recognition system) to get a high performance. But it still has some disadvantages, one of which is that it needs to construct a high dimension of input vector, so it will waste a lot of memory storage and spend much time in computing. In this paper we present a new method to combine HMM and BPNN to decrease the dimension of input vector and still keep a high recognition rate while recognizing.