摘要
根据加权倒谱距离测度,构造了一种用于非特定人语音识别的函数链神经网络,并结合所提出的自适应函数链神经网络学习算法,不但大大缩短了网络的学习时间,而且使得该网络具有较高的识别率,采用10人(5男、5女)的语音数据对汉语10个数字进行了实验,正确识别率为94.2%.
The functional - link neural network for speaker- independent speech recognition is put forward based on weighted cepstral distance measure and self - adaptive functional - link neural network algorithm. The network has not only higher recognition rate but also shortening the training time. An experimental result using 10 numbers, which have been pronounced by 5 males and 5 females, shows that an average recognition rate of 94.2 %.
出处
《邵阳学院学报（自然科学版）》
2006年第2期43-45,共3页
Journal of Shaoyang University：Natural Science Edition
关键词
语音识别
自适应
神经网络
非特定人
Speech recognition
self- adaptive
neural network
speaker - independent
