基于新阈值函数和自适应阈值的小波包语音增强研究

发布时间：2019-02-08 17:10

【摘要】：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。
[Abstract]:In order to solve the serious problem of speech distortion after enhancement by traditional wavelet packet speech enhancement algorithm, a new wavelet packet speech enhancement algorithm based on adaptive threshold and new threshold function is proposed. In the wavelet packet domain, the noisy speech is windowed into frames, and the probability of each frame is calculated based on the cross-correlation value of the power spectrum of the FFT power spectrum of adjacent frames. Then the traditional general wavelet packet threshold is adjusted by speech existence probability, which makes the threshold larger in non-speech frames and smaller in speech frames. The adaptive adjustment of threshold value can eliminate the noise at the same time to the greatest extent. Keep the voice as much as possible and reduce the distortion. A new threshold function is also designed, which overcomes the disadvantages of traditional hard threshold function discontinuity and soft threshold function which will bring constant deviation, and further reduces the speech distortion. The speech and noise in TIMIT and NOISEX-92 database are used in a large number of simulation experiments. The subjective and objective evaluation results show that the proposed speech enhancement algorithm has better enhancement effect than the existing two algorithms. The proposed algorithm has less distortion and better auditory effect.
【作者单位】：安徽师范大学物理与电子信息学院;
【基金】：安徽省科技攻关项目(1501041137) 安徽省教育厅自然科学基金资助项目(KJ2015A079) 安徽师范大学博士科研启动基金资助项目
【分类号】：TN912.35

【相似文献】