基于加权组合过零峰值幅度特征的抗噪语音识别  

A Noise-Robust Speech Recognition Based On Weighting Combinition Zero-crossings with Peak Amplitudes

作　　者：梁五洲[1] 张雪英[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》2006年第1期84-86,共3页Journal of Taiyuan University of Technology

基　　金：国家自然科学基金(60472094);教育部留学回国科研基金(教外司留[2004]176号);山西省留学回国人员科研基金(200224);山西省高等学校青年学术带头人科研基金(晋教科[2004]13号文件);山西省自然科学基金(20051039)

摘　　要：基于人耳听觉特性提出一种新的抗噪音识别特征:加权组合过零峰值幅度特征,是对过零峰值幅度特征的一种改进。加权组合过零峰值幅度特征以语音数据和差分语音数据作为处理对象,通过计算它们的上升过零率获得频率信息,经幅度非线性压缩获得密度信息,并根据人耳对声音的感知特点对其进行加权,形成最终的输出特征,识别网络使用HMM。仿真实现了使用新特征与原特征的算法识别结果,证明了新特征具有较高的识别率和优良的抗噪性能。This paper presents a new approach to extract anti-noisy speech feature： weighting combinition zero-crossings with peak amplitudes, which is based on human auditory model. It is an improved model of ZCPA（Zero-Crossings with Peak Amplitudes）. This approach uses the speech signal and it＇s difference signal as input. The frequency information of speech signal is obtained by upward-going zero-crossing intervals, and the intension information is incorporated by compressed nonlinear amplitudes. The speech feature is weighted based on the auditory characteristics by using weighting function and then the finally feature is outputed. The recognition part uses HMM. Experimental results of speech recognition demonstrate that this new feature is more robust than the older feature in noise environment.

关 键 词：特征提取 过零率 语音识别 

分 类 号：TN912.3[电子电信—通信与信息系统]