基于覆盖的神经网络集成在语音识别中的应用

神经网络集成通过训练多个神经网络并将各网络的结论进行合成,从而得到最终结果.集成可以显著的提高学习系统的泛化能力.讨论了基于覆盖思想而设计的神经网络集成方法,并将其应用于汉语孤立数码语音识别系统中,通过在集成过程中加入基于覆盖思想的控制算法降低系统的泛化误差,从而使系统的识别效果有了进一步的提高.

Application of Neural Network Ensemble Based on Covering Algorithm in Speech Recognition

Neural network ensemble can significantly improve the generalization ability of learning systems through training a finite number of neural networks and then combining their results. It is not only helpful for scientists to investigate machine learning and neural computing but also helpful for common engineers to solve real-world problems using neural network techniques. Therefore neural network ensemble has been regarded as an engineering neural computing technology that has great application prospect. Also it has become a hot topic in both machine learning and neural computing communities. This paper discusses neural network ensemble based on covering algorithm and introduces how to combine multi-results of neural networks and how to create individual network in network ensemble. In order to improve effect of network ensemble and reduce generalization error, the control algorithm based on constructive covering approach will be added in network ensemble. And the detailed algorithm will be given in this paper. After turning, the new network achieves 99.25% and 94.25% correct recognition accuracy without rejection when applied to speaker-dependent and speaker-independent isolated mandarin digit speech recognition. Such performance is much better than single network based on covering algorithm and the network ensemble based on Bagging algorithm.