摘要
实小波包变换是语音增强中效果较好的一种算法,利用阈值的方法对小波包系数进行压缩进而重构语音信号。分析了实小波包变换的平移敏感性,以及其对语音进行增强时的缺陷。提出采用双树复小波包变换方法进行语音增强,当低通滤波器和高通滤波器对应的小波基近似为希尔伯特变换对时,该变换能大大减小实小波包变换中的平移敏感性。同时考虑小波包系数之间的相互关系,提出了重叠块复阈值算法。结果表明,算法优于传统实小波包变换及点阈值算法,尤其对含周期噪声的语音信号,双树复小波包变换算法的优势更为明显。
Real wavelet Packet Transform （RWPT）is a better algorithm for speech enhancement, uses threshold to shrink wavelet packet coefficients and reconstruct speech signal. The property of shift sensitivity of RWPT and its disadvantages in speech enhancement is analysised, Dual-Tree Complex Wavelet Packet Transform （DTCWPT）is proposed to speech enhancement,DTCWPT can reduce shift sensitivity largely when the basis functions of low-pass filter and high-pass filter form Hilbert transform pairs. And the relationship of coefficients of wavelet packet is considered, and overlap block complex threshold algorithm is proposed. The results show the proposed algorithm is better than common RWPT and node threshold method,especially for the speech contained periodic noise.
出处
《传感技术学报》
CAS
CSCD
北大核心
2009年第7期983-987,共5页
Chinese Journal of Sensors and Actuators
关键词
语音增强
复小波
平移不变性
周期噪声
双树复小波包变换
重叠块复阈值
speech enhancement
complex wavelet
shift invariance
periodic noise
Dual-Tree Complex Wavelet Packet Transform（DTCWPT）
overlap block complex threshold
