摘要
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 curve fitting and principalcom ponentanalysis m ethod, this paperpre- sents a novelANN-based speech recognition feature. The feature reflects the variation ofvocaltractw ith tim e. The extraction m ethod sim ulates the processing ofspeech in- form ation in hum an auditory system . Com pared with otherANN-based recognition fea- tures, thisnew featurenotonly increasetherecognition accuracy butalso has follow ing properties: lesscom plex algorithm , lessstoragem em ory and easy realization w ith hard- w are.
出处
《应用科学学报》
CAS
CSCD
1999年第4期427-437,共11页
Journal of Applied Sciences
基金
国家攀登计划认知科学中神经网络理论与应用基础研究重大关键资助
关键词
主分量分析
特征提取
语音识别
语音信号
principalcom ponent analysis, neuralnetwork, feature extraction, speech recognition
