应用科学学报 ›› 1999, Vol. 17 ›› Issue (4): 427-432.

• 论文 • 上一篇    下一篇

语音信号的主分量特征

何振亚, 顾明亮, 王太君, 史笑兴   

  1. 东南大学
  • 收稿日期:1998-03-09 修回日期:1998-09-03 出版日期:1999-12-31 发布日期:1999-12-31
  • 基金资助:
    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项目

Principal Component Feature for Speech Recognition

HE ZHENYA, GU MINGLIANG, WANG TAIJUN, SHI XIAOXING   

  1. Southeast University, Nanjing 210096
  • Received:1998-03-09 Revised:1998-09-03 Online:1999-12-31 Published:1999-12-31

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.

关键词: 主分量分析, 神经网络, 语音识别, 特征提取

Abstract: Using curve fitting and principal component analysis method,this paper presents a novel ANN-based speech recognition feature.The feature reflects the variation of vocal tract with time.The extraction method simulates the processing of speech information in human auditory system.Compared with other ANN-based recognition features,this new feature not only increase the recognition accuracy but also has following properties:less complex algorithm,less storage memory and easy realization with hard ware.

Key words: principal component analysis, feature extraction, speech recognition, neural network