Application of Fuzzy Clustering Neural Network in Conjunction Speech Recognition

In order to improve the approximation property of the past fuzzy clustering algorithms when identifying systems, a fuzzy clustering neural network (FCNN) is proposed and is applied to conjunction speech recognition system. Based on the fuzzy system model, FCNN presents every state as a fuzzy system and uses continuous frames as the system input. With improving fuzzy clustering identification algorithm, FCNN is acted as estimator of probability density function which could forecast output probability of the each state. This model not only can describe the inter-frames correlation information for speech signal efficiently, but overcome the deficiency of traditional hidden markov model which supposes each state’s output is mixed Gauss distributing probability density function. Through the experiments of speakerindependent conjunction speech recognition, the effectiveness of FCNN could be verified.