【正文】 2． 語(yǔ)音大體上可以分為清音和濁音兩大類 濁音在時(shí)域上呈現(xiàn)出明顯的周期性，在頻域上有共振峰結(jié)構(gòu)，而且能量大部分集中在較低頻段內(nèi)；清音則沒(méi)有明顯的時(shí)域和頻域特征，類似于白噪聲。 ???? ??1 )()(1 Nnnmm mnwnxNM （ 26） 但 mM 的動(dòng)態(tài)范圍（最大值與最小值之比）要比短時(shí)能量小，接近于短時(shí)能量計(jì)8 算的平方根，所以用 mM 區(qū)分清音 /濁音、無(wú) /有聲不如短時(shí)能量明顯。在語(yǔ)音信號(hào)數(shù)字處理中，都是采用 nws??? 的 離散傅立葉變換（ DFT） ()Sk? 來(lái)代替(exp( ))Sj? ? ，并且可以用高效的快速傅立葉變換（ FFT） 算法完成由 nws??? 至 ()Sk? 的轉(zhuǎn)換。 加窗處理 通常采用一個(gè)長(zhǎng)度有限的窗函數(shù) w(n)來(lái)乘語(yǔ)音信號(hào) s(n),從而形成加窗語(yǔ)音 )(nsw ＝ s(n)其轉(zhuǎn)移函數(shù)為： )()()()( ZRzVzUzH ? (23) 語(yǔ)音信號(hào)的短時(shí)分析技術(shù) 語(yǔ)音信號(hào)是一種非平穩(wěn)的時(shí)變信號(hào)， 產(chǎn)生過(guò)程與發(fā)音器官的運(yùn)動(dòng)密切相關(guān)，而這種物理運(yùn)動(dòng)比起聲音振動(dòng)速度要緩慢得多，因此語(yǔ)音信號(hào)常常假定為短時(shí)平穩(wěn)的，即在 10～ 30ms 時(shí)間段內(nèi)，其頻譜特性和某些物理特征參量可近似地看作是不變的。 圖 21 語(yǔ)音信號(hào)產(chǎn)生的數(shù)學(xué)模型 激勵(lì)源模型 激勵(lì)源分濁音和清音兩個(gè)分支，按照濁音 /清音開關(guān)所處的位置來(lái)決定產(chǎn)生的語(yǔ)音是濁音還是清音。講話時(shí)，舌和唇連續(xù)運(yùn)動(dòng)，使聲道常常改變外形和尺寸，即改變諧振頻率。 語(yǔ)音的產(chǎn)生 產(chǎn)生語(yǔ)音的能量，來(lái)源于正常呼吸時(shí)肺部呼出的穩(wěn)定氣流。 本文將重點(diǎn)研究和實(shí)現(xiàn)基于譜減法語(yǔ)音增強(qiáng)方法。單通道語(yǔ)音增強(qiáng)是語(yǔ)音增強(qiáng)的基礎(chǔ)。比如將小波變換用于語(yǔ)音增 強(qiáng)中，利用神經(jīng)網(wǎng)絡(luò) (ANN． Arificial Neural Net、 vorks)進(jìn)行語(yǔ)音增強(qiáng)，基于麥克風(fēng)陣列的語(yǔ)音增強(qiáng)等。 （ 3） 在國(guó)家和社會(huì)安全方面，偵聽(tīng)信號(hào)常常含有較大的噪聲，語(yǔ)音增強(qiáng)有助于 提高偵聽(tīng)系統(tǒng)的效果，可以幫助偵察破案或獲取情報(bào)。語(yǔ)音數(shù)字信號(hào)處理包含的內(nèi)容十分 泛 ， 包括語(yǔ)音編碼、語(yǔ)音識(shí)別、語(yǔ)音合成和語(yǔ)音增強(qiáng)等。 本論文 首先從語(yǔ)音特性、噪聲特性、人耳的感知特性以及語(yǔ)音信號(hào)分析得方法入手，重點(diǎn)研究了基于譜減法的增強(qiáng)算法并在 MATLAB 環(huán)境下對(duì)其進(jìn)行了仿真，驗(yàn)證了譜減法 在語(yǔ)音增強(qiáng)方面 的有效性和可行性。 關(guān)鍵詞 ： 語(yǔ)音信號(hào)處理 ,語(yǔ)音增強(qiáng) ,譜減 法 ,DSP II SPEECH ENHANCEMENT ALGORITHMS AND IMPLEMENTATION ON DSP ABSTRACT Speech is inevitably interfered by noise. The noise not only degrades the quality and the intelligibility of speech, but also worsens the capability of the system. As speech enhancement technology can be used to reduce the noise, Improve antinoise ability and the signaltonoise ratio of the input signal , the input SignaltoNoise Ratio of the speech processing system and improving the quality and intelligibility of speech, speech enhancement technology is usually used as the preprocessing module in the speech processing system. This paper first from speech characteristics, noise characteristics, the perception of the human ear characteristics and method of speech signal analysis, Focus on the based on the spectral subtraction enhancement algorithm and in the MATLAB environment of the simulation, Verify the spectral subtraction speech enhancement in the effectiveness and feasibility. DSP as special digital signal processor, has some unique features, such as Singlecycle multiply and accumulate (MAC)， multiple accessing memories in single cycle,a wide variety of onchip peripherals. These features lead DSP applied widely in digital speech processing field. This paper studies the traditional Spectral Subtraction and improves it to reduce “music noise”. Then based on these theories, a practical speech enhancement processing system on DSP is designed and implemented. KEY WORDS: Speech processing,Speech enhancement,Spectral Subtraction,DSP III 基于譜減法的語(yǔ)音增強(qiáng)及其 DSP 實(shí)現(xiàn) ........................................................................................... I 摘 要 ........................................................................................................................................... I ABSTRACT .................................................................................................................................II 第一章 緒論 ................................................................................................................................ 1 課題研究意義 .................................................................................................................. 1 國(guó)內(nèi)外研究現(xiàn)狀及發(fā)展背景 ............................................................................................. 2 主要研究?jī)?nèi)容 .................................................................................................................. 3 第二章 語(yǔ)音增強(qiáng)的基礎(chǔ)知識(shí) ....................................................................................................... 4 ..................................................................................................................... 4 人的聽(tīng)覺(jué)特性 ................................................................................................................ 4 語(yǔ)音信號(hào)產(chǎn)生的數(shù)字模型 ............................................................................................... 5 .......................................................................................................... 5 .............................................................................................................. 5 .............................................................................................................. 5 ................................................................................................ 6 、采樣、 A/D變換 ...................................................................................... 6 .............................................................................................................. 6 .............................................................................................................. 7 ..................................................................................... 7 .......................................................................................................... 8 ........................................................................................................................ 8 ....................................................................................................... 9 .............................................................................................................. 9 .............................................................................................. 10 第三章 譜減法的原理、算法及分析 ............................................................................................11 ................................................................................................................11 譜減法的實(shí)現(xiàn)與仿真 ............................................................................