V. Homomorphic Signal Processing  5-A Homomorphism Homomorphism is a way of “carrying over” operations from one algebra system into another. Ex. convulution multiplication addition 把複雜的運算，變成效能相同但較簡單的運算

 5-B Cepstrum 157 For the system D*[．] convolution product addition  ＋ ＋ * x[n] FT[．] Log[ ] FT-1[．] X(F) FT: discrete-time Fourier transform

 由 y[n]= x [n]*h[n] 重建 x[n] ＋ ＋ ＋ ＋ D*[ ] Linear Filter D-1*[ ] * For the system D-1*[．]  有趣的名詞 cepstrum quefrency lifter

 5-C Complex Cepstrum and Real Cepstrum ambiguity for phase Problems: (1) (2) Actually, the COMPLEX Cepstrum is REAL for real input

160  Real Cepstrum (but some problem of is not solved)

 5-D Methods for Computing the Cepstrum  Method 1: Compute the inverse discrete time Fourier transform: Problem: may be infinite

 Method 2 (From Poles and Zeros of the Z Transform) 162  Method 2 (From Poles and Zeros of the Z Transform) time delay where Poles & zeros inside unite circle outside unite circle

163 Taylor series Z-1 (inverse Z transform) ?

Taylor series expansion 164 Z-1 Taylor series expansion (Suppose that r = 0) Poles & zeros inside unit circle, right-sided sequence Poles & zeros outside unit circle, left-sided sequence Note: (1) 在 complex cepstrum domain Minimum phase 及 maximum phase 之貢獻以 n = 0 為分界切開 (2) For FIR case, ck = 0, dk = 0 (3) The complex cepstrum is unique and of infinite duration for both positive & negative n, even though x[n] is causal & of finite durations

 Method 3 Z-1

166 Suppose that x[n] is causal and has minimum phase, i.e. x[n] = = 0, n < 0   For a minimum phase sequence x[n]

167 For anti-causal and maximum phase sequence, x[n] = = 0, n > 0   For maximum phase sequence,

 5-E Properties P.1 ) The complex cepstrum decays at least as fast as P.2 ) If X(Z) has no poles and zeros outside the unit circle, i.e. x[n] is minimum phase, then     because of no bk, dk P.3 ) If X(Z) has no poles and zeros inside the unit circle, i.e. x[n] is maximum phase, then because of no ak, ck

169 P.4 ) If x[n] is of finite duration, then has infinite duration

 5-F Application of Homomorphic Deconvolution 170  5-F Application of Homomorphic Deconvolution (1) Equalization for Echo Let p[n] be p[n] =δ[n] +αδ[n-Np] x[n] =s[n] +α s[n-Np] =s[n] * p[n] Z-1

Filtering out the echo by the following “lifter”: 171 Filtering out the echo by the following “lifter”: (2) Representation of acoustic engineering x[n] = s[n] * h[n] Np 2Np 3Np Signal n Synthesized music music building effect：eg. 羅馬大教堂的impulse response

172 (3) Speech analysis (4) Seismic Signals (5) Multiple-path problem analysis Speech wave Pitch Global wave shape Vocal tract impulse They can be separated by filtering in the complex Cepstrum Domain

173 用 cepstrum 將 pitch 的影響去除 From 王小川， “語音訊號處理”，全華出版，台北，民國94年。

174

 5-G Problems of Cepstrum 175  5-G Problems of Cepstrum (1) | log(X(Z))| (2) Phase (3) Delay Z-k (4) Only suitable for the multiple-path-like problem

 5-H Differential Cepstrum 或 inverse Z transform Note: If Advantages: no phase ambiguity able to deal with the delay problem

 Properties of Differential Cepstrum (1) The differential Cepstrum is shift & scaling invariant 不只適用於 multi-path-like problem 也適用於 pattern recognition If y[n] = A X[n - r]   (Proof):

(2) The complex cepstrum is closely related to its differential cepstrum and the signal original sequence x[n]   Complex cepstrum 做得到的事情, differential cepstrum 也做得到！

(5) If x(n) is of finite duration, has infinite duration 179 (3) If x[n] is minimum phase (no poles & zeros outside the unit circle), then = 0 for n  0 (4) If x[n] is maximum phase (no poles & zeros inside the unit circle) , then = 0 for n  2 1 2 delay max phase min phase (5) If x(n) is of finite duration, has infinite duration Complex cepstrum decay rate Differential Cepstrum decay rate 變慢了,  

 5-I Mel-Frequency Cepstrum (梅爾頻率倒頻譜) 180  5-I Mel-Frequency Cepstrum (梅爾頻率倒頻譜) Take log in the frequency mask Bm[k] = 0 for k < fm1 and k > fm+1 for fm1  f  fm for fm  f  fm+1 gain mask of Mel-frequency cepstrum frequency fm-1 fm fm+1

summation of the effect inside the mth mask 181 Process of the Mel-Cepstrum summation of the effect inside the mth mask x[n] X[k] Q: What are the difference between the Mel-cepstrum and the original cepstrum? Mel-frequency cepstrum 更接近人耳對語音的區別性   用 cx[1], cx[2], cx[3], ……., cx[13] 即足以描述語音特徵 FT

182  5-J References  R. B. Randall and J. Hee, “Cepstrum analysis,” Wireless World., vol. 88, pp. 77-80. Feb. 1982  王小川， “語音訊號處理”，全華出版，台北，民國94年。  A. V. Oppenheim and R. W. Schafer, Discrete-Time Signal Processing, London: Prentice-Hall, 3rd ed., 2010.  S. C. Pei and S. T. Lu, “Design of minimum phase and FIR digital filters by differential cepstrum,” IEEE Trans. Circuits Syst. I, vol. 33, no. 5, pp. 570- 576, May 1986.  S. Imai, “Cepstrum analysis synthesis on the Mel-frequency scale,” ICASSP, vol. 8, pp. 93-96, Apr. 1983.

附錄六：聲音檔和影像檔的處理 (by Matlab) 183 附錄六：聲音檔和影像檔的處理 (by Matlab) A. 讀取聲音檔 電腦中，沒有經過壓縮的聲音檔都是 *.wav 的型態 讀取： wavread 例： [x, fs] = wavread('C:\WINDOWS\Media\ringin.wav'); 可以將 ringin.wav 以數字向量 x 來呈現。 fs: sampling frequency 這個例子當中 size(x) = 9981 1 fs = 11025 思考: 所以，取樣間隔多大? 這個聲音檔有多少秒？

184 畫出聲音的波型 time = [0:length(x)-1]/fs; % x 是前頁用 wavread 所讀出的向量 plot(time, x) 注意： *.wav 檔中所讀取的資料，值都在 1 和 +1 之間

185 一個聲音檔如果太大，我們也可以只讀取它部分的點 [x, fs]=wavread('C:\WINDOWS\Media\ringin.wav', [4001 5000]); % 讀取第4001至5000點 [x, fs, nbits] = wavread('C:\WINDOWS\Media\ringin.wav'); nbits: x(n) 的bit 數 第一個bit : 正負號，第二個bit : 21，第三個bit : 22， …..， 第 n 個bit : 2nbits +1， 所以 x 乘上2nbits 1 是一個整數 以鈴聲的例子， nbits = 8，所以 x 乘上 128是個整數

186 有些聲音檔是 雙聲道 （Stereo）的型態 (俗稱立體聲) 例： [x, fs]=wavread('C:\WINDOWS\Media\notify.wav'); size(x) = 29823 2 fs = 22050

187 B. 繪出頻譜 (請參考附錄四) X = fft(x); plot(abs(X)) fft 橫軸 轉換的方法 (1) Using normalized frequency F: F = m / N. (2) Using frequency f, f = F  fs = m  (fs / N).

188

189 C. 聲音的播放 (1) wavplay(x): 將 x 以 11025Hz 的頻率播放 (時間間隔 = 1/11025 = 9.07  105 秒) (2) sound(x): 將 x 以 8192Hz 的頻率播放 (3) wavplay(x, fs) 或 sound(x, fs): 將 x 以 fs Hz 的頻率播放 Note: (1)~(3) 中 x 必需是1 個column (或2個 columns)，且 x 的值應該 介於 1 和 +1 之間 (4) soundsc(x, fs): 自動把 x 的值調到 1 和 +1 之間 再播放

190 D. 用 Matlab 製作 *.wav 檔： wavwrite wavwrite(x, fs, waveFile) 將數據 x 變成一個 *.wav 檔，取樣速率為 fs Hz  x 必需是1 個column (或2個 columns)  x 值應該 介於 1 和 +1 之間  若沒有設定fs，則預設的fs 為 8000Hz

範例程式： 191 E. 用 Matlab 錄音的方法 錄音之前，要先將電腦接上麥克風，且確定電腦有音效卡 (部分的 notebooks 不需裝麥克風即可錄音) 範例程式： Sec = 3; Fs = 8000; recorder = audiorecorder(Fs, 16, 1); recordblocking(recorder, Sec); audioarray = getaudiodata(recorder); 執行以上的程式，即可錄音。 錄音的時間為三秒，sampling frequency 為 8000 Hz 錄音結果為 audioarray，是一個 column vector (如果是雙聲道，則是兩個 column vectors)

範例程式 (續)： 192 wavplay(audioarray, Fs); % 播放錄音的結果 t = [0:length(audioarray)-1]./Fs; plot (t, audioarray‘); % 將錄音的結果用圖畫出來 xlabel('sec','FontSize',16); wavwrite(audioarray, Fs, ‘test.wav’) % 將錄音的結果存成 *.wav 檔

指令說明： 193 recorder = audiorecorder(Fs, nb, nch); (提供錄音相關的參數) Fs: sampling frequency, nb: using nb bits to record each data nch: number of channels (1 or 2) recordblocking(recorder, Sec); (錄音的指令) recorder: the parameters obtained by the command “audiorecorder” Sec: the time length for recording audioarray = getaudiodata(recorder); (將錄音的結果，變成 audioarray 這個 column vector，如果是雙聲道，則 audioarray 是兩個 column vectors) 以上這三個指令，要並用，才可以錄音

194 F：影像檔的處理 Image 檔讀取: imread Image 檔顯示: imshow, image, imagesc Image 檔製作: imwrite 基本概念：灰階影像在 Matlab 當中是一個矩陣 彩色影像在 Matlab 當中是三個矩陣，分別代表 Red, Green, Blue *.bmp: 沒有經過任何壓縮處理的圖檔 *.jpg: 有經過 JPEG 壓縮的圖檔 Video 檔讀取: aviread

範例一： (黑白影像) 範例二：(彩色影像) 195 im=imread('C:\Program Files\MATLAB\pic\Pepper.bmp'); (注意，如果 Pepper.bmp 是個灰階圖，im 將是一個矩陣) size(im) ans = 256 256 (用 size 這個指令來看 im 這個矩陣的大小) image(im); colormap(gray(256)) 範例二：(彩色影像) im2=imread('C:\Program Files\MATLAB\pic\Pepper512c.bmp'); size(im2) ans = 512 512 3 (注意，由於這個圖檔是個彩色的，所以 im2 將由三個矩陣複合而成) imshow(im);