MDSR Based on Fuzzy Clustering Neural Network

In order to overcome inherent bugs of basic hidden markov model (HMM), a method of speech recognition based on fuzzy clustering neural network is presented. Based on the fuzzy system model, every state (HMM) is regarded as a fuzzy system in this method. With continuous frames character vector of speech signal as the system´s input, the model can forecast the probability density function of the system´s output states by using improved fuzzy clustering identifying algorithm to build a novel fuzzy clustering neural network. It not only can import the relativity of frames about speech signal efficiently, it also can overcome the limit chain of mixed Gauss distributing probability density function. Speaker independent mandarin digit speech recognition which based on this method is realized. Experimental results show that the method is efficiency and has higher recognition ratio than basic HMM.