CAJ | 학술논문

利用小波多分辨率的理论对语音信号进行信号分解,结合其发声特性,分析高低频段对说话人识别的贡献大小,根据识别结果的分析,提取出了可以综合识别时间和识别效率的特征参数.实验结果表明,一层分解后的小波细节系数识别率为94.4%,比原信号 MFCC 提高7%,而数据个数却比原信号降低了一半,二次分解后的高频段语音依然得到了较高识别率,提取出的较低频信号也可以达到70.8%的识别率.
이용소파다분변솔적이론대어음신호진행신호분해,결합기발성특성,분석고저빈단대설화인식별적공헌대소,근거식별결과적분석,제취출료가이종합식별시간화식별효솔적특정삼수.실험결과표명,일층분해후적소파세절계수식별솔위94.4%,비원신호 MFCC 제고7%,이수거개수각비원신호강저료일반,이차분해후적고빈단어음의연득도료교고식별솔,제취출적교저빈신호야가이체도70.8%적식별솔.
This paper applies the theory of wavelet multiresolution theory to decomposing the signal. Combining with its voice features, this paper analyses the difference of the contribution of the high and low frequency on the speaker recognition. Experi-mental results show that the rate of detail coefficients after a layer of decomposition is 94.4%, increased by 7% than the MFCC of the original signal, and the number of the data cuts by half than the original signal. The second dissolved high frequencies voice still achieves higher recognition rate, and the extracted low-frequency signal also can achieve 70.8% recognition rate.