论文部分内容阅读
在传统的ＨＭＭ语音识别方法的基础上，提出了两种改进的竞争神经网络算法，分别用于语音识别的两个不同方面．首先提出了一种基于选择机制的新的竞争算法，这种算法可以有目的性地避免局部最优，而且可以克服模拟退火算法（ＳＡ）的随机性．然后，针对分类器的特性，对竞争算法进行改进，把安全拒识措施结合到竞争算法中，提出了一种新颖的神经网络——并行、自组织、层次神经网（ＰＳＨＮＮ）．实验结果表明，基于竞争神经网络算法的语音识别系统比传统的语音识别系统在识别能力和识别速度上都有明显提高，从而证明了与竞争神经网络算法结合的语音识别方法是可行的，而且具有良好的发展和应用前景．
Based on the traditional HMM speech recognition method, two improved competitive neural network algorithms are proposed, which are respectively used for two different aspects of speech recognition. Firstly, a new competition algorithm based on the selection mechanism is proposed. This algorithm can avoid local optimization purposely and overcome the randomness of the simulated annealing algorithm (SA). Then, aiming at the characteristics of the classifier, the competition algorithm is improved and the security rejection method is incorporated into the competition algorithm. A novel neural network - parallel, self-organizing and hierarchical neural network (PSHNN) is proposed. The experimental results show that the speech recognition system based on competitive neural network algorithm has a significant improvement in recognition ability and recognition speed compared with the traditional speech recognition system, which proves that the speech recognition method combined with competitive neural network algorithm is feasible, and has Good development and application prospects.