鄭珩+宋斌
摘 要: 語(yǔ)音信號(hào)降噪是語(yǔ)音信號(hào)處理的重要方面,對(duì)人們的日常生活有重要影響。常見(jiàn)的降噪算法包括自適應(yīng)方法及小波變換等。LMS算法是自適應(yīng)算法的一種,其優(yōu)點(diǎn)是算法實(shí)現(xiàn)簡(jiǎn)單,復(fù)雜度低;其主要缺點(diǎn)是不能同時(shí)保證算法的收斂速度和實(shí)驗(yàn)精度。小波閾值在處理語(yǔ)音信號(hào)方面也有獨(dú)特的優(yōu)勢(shì),其主要缺點(diǎn)是處理高頻信號(hào)時(shí)存在失真現(xiàn)象。提出將兩種方法結(jié)合起來(lái),對(duì)語(yǔ)音信號(hào)進(jìn)行兩次處理,保留了每種算法的優(yōu)點(diǎn),同時(shí)避免單一方法帶來(lái)的不利影響。實(shí)驗(yàn)結(jié)果表明,該方法的效果顯著優(yōu)于單一算法。
關(guān)鍵詞: LMS; 小波閾值; 語(yǔ)音降噪; 信號(hào)處理
中圖分類(lèi)號(hào): TN911?34; TP301.6 文獻(xiàn)標(biāo)識(shí)碼: A 文章編號(hào): 1004?373X(2015)08?0037?04
Research on voice noise reduction algorithm based on LMS algorithm and wavelet threshold
ZHENG Hang, SONG Bin
(School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210014, China)
Abstract: Noise reduction is an important aspect of the voice signal processing, which has a major impact on people's daily lives. The common methods of noise reduction include adaptive noise reduction algorithm and wavelet transform algorithm. LMS algorithm is one of the adaptive algorithms, whose advantages are low complexity and easy to realize, but whose main drawbacks are those which can not guarantee the convergence rate of the algorithm and the accuracy of the experimental at the same time. Wavelet threshold algorithm also has a unique advantage in dealing with the voice signal, whose main drawback is distortion when dealing with high?frequency signals. The combination of two methods to deal with voice signal twice to retain the advantages of each algorithm and avoid the influence caused by a single method is proposed in this paper. The experimental results show that the effect of the method is significantly better than only one algorithm.
Keywords: LMS; wavelet threshold; voice noise reduction; signal processing
0 引 言
聲音是人與人之間溝通交往的一種重要方式。然而在特定環(huán)境下會(huì)產(chǎn)生噪聲,輕則造成聲音聽(tīng)起來(lái)比較刺耳,重則無(wú)法獲取聲音中的有效信息。如何減少甚至消除噪聲是現(xiàn)在一個(gè)廣泛研究的課題。
現(xiàn)在常用的噪聲消除方法包括譜減法、線性濾波法、小波變換法,子空間語(yǔ)音降噪法,自適應(yīng)噪聲抵消法[1]等。其中自適應(yīng)噪聲抵消法[2?3]效果較好,相比于其他方法,由于多了一個(gè)參考噪聲作為輸入,因此能夠獲得更加全面的噪聲信息,從而得到更加良好的降噪效果。LMS作為自適應(yīng)噪聲抵消方法的一種,具有計(jì)算量較少的有點(diǎn)。另外小波分析[4]是一種時(shí)域分析,克服了傅里葉變換固定分辨率的缺點(diǎn),在信號(hào)的高頻部分,可以獲得較好的時(shí)間分辨率,在信號(hào)的低頻部分可以獲得較高的頻率分辨率,適用于語(yǔ)音信號(hào)等非平穩(wěn)信號(hào)的處理。
1 LMS自適應(yīng)算法
1.1 LMS算法原理
自適應(yīng)濾波[5]就是利用前一時(shí)刻獲得濾波器參數(shù)的結(jié)果自動(dòng)的調(diào)節(jié)現(xiàn)時(shí)刻的濾波器參數(shù),以適應(yīng)信號(hào)和噪聲統(tǒng)計(jì)特性,從而實(shí)現(xiàn)最優(yōu)濾波。自適應(yīng)濾波器實(shí)質(zhì)上就是一種能調(diào)節(jié)其自身傳輸特性以達(dá)到最優(yōu)的濾波器。一般自適應(yīng)濾波器的結(jié)構(gòu)如圖1所示。
圖1中[x(n)]為輸入信號(hào),通過(guò)參數(shù)可調(diào)的數(shù)字濾波器后產(chǎn)生輸出信號(hào)[y(n)],將輸出信號(hào)[y(n)]與期望信號(hào)[d(n)]進(jìn)行比較,得到誤差信號(hào)[ε(n)]。[ε(n)]和[x(n)]通過(guò)自適應(yīng)算法對(duì)濾波器參數(shù)進(jìn)行調(diào)整,使誤差信號(hào)[ε(n)]最小。