论文部分内容阅读
利用曲线拟合与主分量分析神经网络相结合的方法，提出了一种既反映声道变化规律又符合人耳听觉特点的语音识别新特征． 与其他神经网络识别特征相比，新特征不仅可以提高语音识别准确率，而且具有算法简单、存储容量小、便于实时实现的特点．
Using the method of curve fitting and principal component analysis neural network, a new speech recognition feature that reflects both the variation of the sound channel and the hearing characteristics of the human ear is proposed. Compared with other neural network recognition features, the new feature can not only improve the accuracy of speech recognition, but also has the advantages of simple algorithm, small storage capacity and real-time implementation.