针对传统的小波包语音增强算法增强后的语音失真严重的问题,提出了一种基于自适应阈值和新阈值函数的小波包语音增强算法.该算法在小波包域将带噪语音加窗分帧,基于相邻帧快速傅里叶变换功率谱的互相关值,计算各帧存在语音的概率;然后通过语音存在概率对传统通用小波包阈值进行调整,使得阈值在非语音帧中较大,在语音帧中较小,实现阈值的自适应调整,可以在最大程度消除噪声的同时,尽可能地保留语音,减小语音失真.还设计了一种新阈值函数,克服了传统硬阈值函数不连续和软阈值函数会带来恒定偏差的缺点,进一步减小了语音失真.采用TIMIT和NOISEX-92数据库中的语音和噪声进行了大量的模拟实验,主观评比和客观评比结果均证明提出的语音增强算法比现有的两种算法有更好的增强效果,采用提出的算法增强后的语音失真更小,听觉效果更好.%Aiming at overcoming the drawbacks that speech enhanced by traditional wavelet packet speech enhancement method had severe distortion,this paper proposed a wavelet packet speech enhancement method based on adaptive threshold and a novel thresholding function.The proposed method windowed and framed noisy speech in wavelet packet domain.Then it calculated probability of existance of speech in each frame based on cross-correlation values of the fast Fourier transform power spectrum of two adjacent frames.Finally it obtained adaptive threshold by adjusting traditional universal threshold accroding to the probability,the adaptive threshold was small in speech distributed frames and large in the speech absence frames,which could preserve speech as far as possible at the same time of removing noise in a greater degree,thus reduced speech distortion.This paper also designed a novel thresholding function,which overcame the drawbacks that traditional hard thresholding function was discontinuous and soft thresholding function brought fixed bias in final speech,to further reduce speech distortion.Extensive simulations were carried out using the TIMIT and NOISEX-92 databases to evaluate the effectiveness of the proposed method.And both subjective and objective evaluations prove that the proposed method has a better performance in enhancing noisy speech,and speech enhanced by proposed method has less speech distortion and better auditory effect.
展开▼
