Research On Speaker-Independent Continuous Speech Recognition Based On Biomimetic Pattern Recognition

In this paper, a method of feature extraction combines with LPCC and peak characteristics based on LPCC feature cxtraction. This feature extraction algorithm resolves the disadvantage of computational complexity of MFCC. According to this method build a speech recognition sample library. In the past, speech recognition methods are based on the traditional pattern recognition ones, which only focus on the"distinction", but not the"cognition". In this paper, the theories of biomimetic pattern recognition and high-dimension space covering are applied into the isolated word speech recognition. And, based on Hopfield network and RBF network, a new type of neural network model is constructed to realize the coverage of different types of samples which form different geometrical shapes in high-dimension space. Therefore, the purpose of classification will be well achieved by adopting the above model. The recognition results show that the learning method mentioned in this paper better than the HMM method.