A speech recognition method based clustering neural network integration

An improved BP neural network classifier integration method was mainly described, by which using k-means clustering a group of value of weights and thresholds with some differences were gotten, and then as the value of individuals of integrated network to improve the performance of integrated learning, and be successfully applied to non-specific human isolated word speech recognition system. By comparing the experimental result and the traditional Adaboost integration algorithm, the validity of the method was confirmed.