一种基于听觉模型的抗噪语音识别特征提取方法  

A Noise-Robust Feature Extraction Method Based on Auditory Model in Speech Recognition

作　　者：焦志平[1] 张雪英[1] 赵姝彦[1] 

机构地区：[1]太原理工大学信息工程学院,山西太原030024

出　　处：《太原理工大学学报》2005年第1期13-15,22,共4页Journal of Taiyuan University of Technology

基　　金：国家自然科学基金资助项目(60472094);教育部留学回国科研启动基金资助项目(教外司留[2004]176号文件);山西省留学回国人员科研基金资助项目(200224)

摘　　要：提出了一种新的语音识别特征提取方法。该方法是建立在听觉模型的基础上,通过计算语音的上升过零率作为频率信息并通过非线性幅度加权相结合来获取语音特征。仿真实现了中小词汇量、孤立词的语音识别,得到了较好的实验结果,证明了此方法具有较强的抗噪声性能。This paper introduces a new feature extraction method which is based on auditory model. The frequency information of speech signal is obtained by zero-crossing intervals, and amplitude information is incorporated by nonlinear transform. A medium and small vocabulary isolated word speech recognition system is simulated. The results show that this method has better robustness than traditional methods.

关 键 词：语音识别 特征提取 听觉模型 

分 类 号：TN912[电子电信—通信与信息系统]