汉语普通话平均频谱  

The average spectrum of Chinese speech

作　　者：杨伟[1] 沈勇[1] 刘紫赟[1] 

机构地区：[1]近代声学教育部重点实验室,南京大学声学研究所,南京210093

出　　处：《南京大学学报（自然科学版）》2012年第1期1-7,共7页Journal of Nanjing University（Natural Science）

基　　金：国家自然科学基金(10774075)

摘　　要：通过改变测量设备及数据处理方法、测量环境、语声阅读材料的选择方法以及语声样本发声人的选择方法等,使用语言噪声的测量法,对32个发声人的汉语普通话语声样本进行录制采集分析计算,得到了不同的汉语普通话的平均频谱曲线.语噪声方式比别的方式有着一定的优越性.在南京大学消声室中进行样本采集以提高测量的信噪比.使用多通道数据采集前端录制每个人的语声样本,使用软件分析得到每个人普通话语声的长期声压频谱,通过主观评测与客观评测的方法剔除异常数据,使用数学方法将每个人的长期声压频谱进行自归一化,然后将归一化后的样本进行混录,最终得到不同于已有汉语普通话的平均频谱.平均频谱可作为语言的产生、传输、接收及语言信号处理等语言通信系统、电声器件和测试设备的设计依据,并可作为语言扩声系统予加重的依据和典型的观众噪声谱.New power spectral density level of Chinese speech was obtained in this paper by using the method of speech noise,a method found to be superior in many respects.The speech noise was the phenomena produced by many people speaking together.Based on existing methods for measuring and calculating the average spectrum of a language,this paper adopted some improvements in the following respects: measurement methods,measurement system,experiment environment,speech materials,experiment subjects and calculation methods.In this paper,every Chinese speech material selected typically reflects characteristics of Chinese speech.We chose 32 subjects(16 male and 16 female) for the experiment,each of them has obtained Putonghua Proficiency Rating Certificate,and their ages are in the range from 20～30.Every subject's Chinese speech samples were recorded by multichannel measurement system,and finally 160 Chinese speech samples were obtained.The experiment was conducted in the anechoic room of Nanjing University to improve the signal to noise ratio.In the application range of the anechoic room we used,the anechoic room is basically free field.Before each experiment,the instruments were calibrated.Every subject's long-term spectrum of the sound pressure was measured by the multi-functional software using fast Fourier transform method with 1/3 octave.Subjective evaluation and objective evaluation were used for excluding abnormal data,5 samples were excluded finally.According to the 155 valid speech samples,the average spectrum of Chinese speech was achieved by the calculation process mentioned below: normalized every subject's long-term sound pressure spectrum,mixed all subjects' normalized sound pressure spectrum,normalized the mixed samples as the long-term average overall sound pressure level is 0 dB,and finally calculated the power spectral density level of Chinese speech.The result showed the main energy of Chinese speech spreading in the frequency range of 63～20 kHz.The average spectrum of Chinese speech provides the basis f

关 键 词：平均频谱 汉语普通话 功率谱密度级 

分 类 号：TN912.3[电子电信—通信与信息系统]