摘要
提出一种新的网络结构，这种网络能够很好地解决神经网络语音识别中的时间规整问题．该网络从输入语音信号的特征矢量序列中提取一组固定数目的特征矢量，然后将这组特征矢量馈入神经网络分类器进行识别．和其他的神经网络语音识别方法相比较，用这种网络进行前端处理，可以缩短后端神经网络分类器的训练和识别时间，简化分类器的网络结构并保持较高的识别率．
This paper proposed a novel network structure to solve the time alignment problem in artificial neural network (ANN) based speech recognition. By using this network, a fixed number of feature vectors is extracted from input speech signal, and then recognized by a neural network classifier. Compared with other ANN based methods, our method has many merits such as much faster training and recognition, simpler structure and higher accuracy. A word recognition system is established based on this method, and we test it with two sets of English words. The experimental results demonstrate that the proposed method has the merits stated above.
出处
《东南大学学报（自然科学版）》
EI
CAS
CSCD
1999年第5期47-51,共5页
Journal of Southeast University：Natural Science Edition
基金
国家自然科学基金
关键词
时间规整算法
神经网络
语音识别
time wrapping algorithm
neural network
speech recognition
