摘 要: 針對寬帶噪聲干擾的語音降噪問題,提出一種基于LMS自適應(yīng)噪聲抵消和小波閾值的語音降噪算法。該算法首先采用LMS自適應(yīng)噪聲抵消器對消部分噪聲,得到較高信噪比的語音信號(hào)后,再對其進(jìn)行小波分析,用一種新的閾值函數(shù)對信號(hào)進(jìn)行降噪,再重構(gòu)得到降噪后的信號(hào)。Matlab仿真實(shí)驗(yàn)證明,該算法的效果優(yōu)于單一算法,且避免了傳統(tǒng)譜減法帶來的“音樂噪聲”,視覺效果、輸出信噪比和均方根誤差也有很大改善。
關(guān)鍵詞: LMS; 自適應(yīng)噪聲抵消; 小波閾值; SNR; RMSE
中圖分類號(hào): TN911?34 文獻(xiàn)標(biāo)識(shí)碼: A 文章編號(hào): 1004?373X(2016)03?0027?03
Research on speech denoising algorithm based on LMS adaptive
noise cancellation and wavelet threshold
KE Shuixia, LI Chisheng
(School of Information Engineering, Nanchang University, Nanchang 330031, China)
Abstract: For the speech denoising disturbed with the broadband noise, a speech denoising algorithm based on LMS adaptive noise cancellation and wavelet threshold is presented. The part noise is cancelled by LMS adaptive noise canceller used in the algorithm to obtain the speech signal with higher signal to noise ratio (SNR), and then the wavelet analysis for the signal is conducted. A new threshold function is used to denoise the single, and the denoised single is obtained by reconstruction. The experiment results of Matlab simulation show that the algorithm is better than single algorithm, and can avoid the “musical noise” caused by the traditional spectral subtraction. The visual effect, output SNR and root mean square error (RMSE) were improved greatly.
Keywords: LMS; adaptive noise cancellation; wavelet threshold; SNR; RMSE
0 引 言
在實(shí)際的工程應(yīng)用中,語音信號(hào)不可避免地受到各種噪聲的影響,尤其是與語音信號(hào)在時(shí)域和頻域上完全重疊的寬帶噪聲。為了改善語音質(zhì)量,提高語音的可懂度,國內(nèi)外人士已取得很多研究成果[1],但目前沒有一種方法能夠完全消除噪聲,因而基于各種降噪算法的原理,提出了很多綜合的算法[2?8],都能取得更好的降噪效果。本文在文獻(xiàn)[5?8]的基礎(chǔ)上,提出一種基于LMS自適應(yīng)噪聲抵消和小波閾值的語音降噪算法,同時(shí)改進(jìn)了閾值函數(shù),經(jīng)采用疊加高斯噪聲的語音信號(hào)仿真表明,該算法既避免了傳統(tǒng)譜減法帶來的“音樂噪聲”,計(jì)算又簡單,也在視覺效果、信噪比和均方誤差方面取得有效改善。
1 LMS自適應(yīng)噪聲抵消原理[9?11]
自適應(yīng)噪聲抵消比其他降噪算法多采用了一個(gè)參考噪聲作為輔助輸入,因而降噪效果較好,特別在輔助的輸入噪聲與語音中的噪聲完全相關(guān)的情況下,能徹底抵消語音中的噪聲。……