Comparison of Biomimetic Pattern Recognition,HMM and DTW for Speaker-Independent Speech Recognition
Shoujue Wang · Dianzi xuebao · 2005
AbstrcatThe purpose of this paper is to compare the performance of three speech recognition methods,one based on Biomimetic Pattern Recognition (BPR) and the other two based on Hidden Markov Models (HMMs) and Dynamic Time Warping (DTW) respectively.As a general purpose model of pattern Recognition,BPR is realized by Multi-Weights Neuron Networks.For the 15 words vocabulary,we analyze the false recognition rate (ratio of accepting a trained word to another trained word) and false acceptance rate (ratio of accepting an untrained word to a trained word) respectively.Experiment results show that when the training data was not sufficient,the manner of BPR achieved a higher performance.