基于ELM和PCA的汉语数字语音识别研究

Research of Chinese Digital Speech Recognition Based on ELM and PCA

针对传统BP网络在语音识别应用中存在训练时间长,容易陷入局部极小值等问题,建立了一种基于ELM的语音识别系统。ELM是一种快速的单隐层前馈神经网络(SLFN)训练算法,将该算法与单隐层BP网络进行实验比较。实验中对提取的特征矩阵采用主成分分析(PCA)算法进行降维,该算法有效地提取了语音信号的主要成分。实验结果表明:在训练时间上,ELM明显优于BP算法;在识别率上,ELM优于BP算法。

In order to solve the traditional BP neural network problems like the long training time and easy to fall into local minimum in speech recognition applications. A speech recognition system based on ELM is established in this paper. ELM is a fast training algorithm of single - hidden layer feed forward neural network. The algorithm is compared with single - hidden layer BP neural network through experiments. Principal component analysis algorithm is used to reduce the dimensions of the extracted feature matrix in the experiment, which extracts the main components of the speech signal effectively. The experi- mental results show that the ELM algorithm is much better than BP algorithm in the aspect of training time, and it is better than BP algorithm in the aspect of recognition rate.