基于小波变换和线性预测的基音提取  

Pitch Extraction Based on Wavelet Transformation and Linear Prediction

作　　者：李坤[1] 刘加[1,2] 

机构地区：[1]清华大学电子工程系,北京100084 [2]清华信息科学与技术国家实验室(筹),北京100084

出　　处：《计算机工程》2010年第10期276-278,287,共4页Computer Engineering

基　　金：国家自然科学基金委员会与微软亚洲研究院联合基金资助项目(60776800);国家"863"计划基金资助项目(2006AA010101;2007AA04Z223;2008AA02Z414)

摘　　要：为有效提取基音周期,提出一种改进的小波变换方法,在小波滤波器的基础上,加入倍频抑制系数,以去除大部分的倍频干扰。根据语音情况融合阶数可调的线性预测,能去除强共振峰造成的倍频干扰,同时避免基音信号受到过多损伤。实验结果表明,在安静的环境下,该方法的准确率可达97.8%;在信噪比为0 dB时,其准确率为91.1%,从而使系统具有较好的准确性、鲁棒性和抗噪性。To effectively extract the pitch period,an improved wavelet transformation method is presented.On basis of wavelet filter,the frequency multiplication suppressed coefficients are added to remove most of the frequency multiplication errors.According to the voice,the system combines the linear prediction,which has a suitable order adjusting to the speech,to remove the format interference,while avoiding excessive damage to the pitch.Experimental results show the accuracy rate of the method is 97.8% at quite environment,and when the SNR is 0 dB,the one of it is 91.1%.The system has a good accuracy,robustness,and noise immunity.

关 键 词：基音提取 小波变换 线性预测 

分 类 号：N945[自然科学总论—系统科学]