All the files and source codes for the Case Study 1

aieask · Mar 11, 2022 · c8b6cfe · c8b6cfe
1 parent 7c40f9b
commit c8b6cfe
Show file tree

Hide file tree

Showing 52 changed files with 645 additions and 0 deletions.
diff --git a/casestudy1/19BEC001MLSCVA01.pdf b/casestudy1/19BEC001MLSCVA01.pdf
diff --git a/casestudy1/mlscv_casestudy1.pdf b/casestudy1/mlscv_casestudy1.pdf
diff --git a/casestudy1/question1/STEn_STM_calc.m b/casestudy1/question1/STEn_STM_calc.m
@@ -0,0 +1,26 @@
+function [STEn, STMag] = STEn_STM_calc(speech, WL_samp, WS_Samp, fs)
+
+
+
+Num_frames = round(  (length(speech)-WL_samp)/WS_Samp );
+STEn = zeros(1,Num_frames);
+STMag = zeros(1,Num_frames);
+
+for i=1:Num_frames
+    startsamp = (i-1)*WS_Samp+1;
+    EndSamp = startsamp+WL_samp-1;
+    CurrSpeech  = speech(startsamp :EndSamp);
+    CurrSpeechWin = (CurrSpeech).*hamming(WL_samp);
+    STEn(i) = sum(CurrSpeechWin.*CurrSpeechWin);
+    STMag(i) = sum(abs(CurrSpeechWin));
+
+
+%     if(i==11)
+%         
+%         FFTSpec = fft(CurrSpeechWin, 512);
+%         freq = (0:256)*(fs/512);
+%         figure; plot (freq, 20*log10( abs(FFTSpec(1:257))),'LineWidth',2);
+%         temp=1;
+%     end
+end
+
diff --git a/casestudy1/question1/abhishek_record.wav b/casestudy1/question1/abhishek_record.wav
diff --git a/casestudy1/question1/calc_MFCC.m b/casestudy1/question1/calc_MFCC.m
@@ -0,0 +1,68 @@
+function [MFCC, logFiltBankOut, logFiltBankFreq] = calc_MFCC(speech, N_filters, N_FFT, fs)
+% calculate the MFCC now.
+fmax = fs/2;
+Mel_fmax = 1127*log(1+ (fmax/700));
+
+% obtain the center frequencies in mel scale
+N = N_filters;
+% N_FFT = 512;
+m=0:(N+1);
+c = m*(Mel_fmax/(N+1));
+
+% convert these to linear frequencies
+h = 700* (exp(c/1127) - 1);
+
+% convert to DFT indices
+g = round( (N_FFT*h)/fs) + 1; % +1 added due to MatLab indexing 1
+
+% Obtain the filterbank coefficients
+temp=1;
+
+H = zeros(N, (N_FFT/2)+1); 
+for m=2:N+1
+    for k=1:(N_FFT/2)+1
+        if( k< g(m-1) || k>g(m+1)) 
+            H(m-1, k) = 0;
+        else
+            if(k>=g(m-1) && k<=g(m))
+                H(m-1, k) = (k-g(m-1))/(g(m)-g(m-1));
+            else
+                H(m-1, k) = (g(m+1)-k)/(g(m+1)-g(m));
+            end
+
+        end
+    end
+end
+
+
+%%%%%%%%%%%%calculating MFCC %%%%%%%%%%
+
+spec = abs(  fft(speech.*hamming(length(speech)), N_FFT));
+
+% Filter the spectrum using 
+spec = spec( 1: (N_FFT/2)+1);
+
+% size(spec)
+% size(H)
+Fout = H* spec;
+logFiltBankOut = log(Fout);
+logFiltBankFreq = g(2:N+1) * (fs/N_FFT);
+
+% DCT calculation
+
+for l=1:13
+    DCTout_temp=0;
+    for m=1:N
+        DCTout_temp  = DCTout_temp+ log(Fout(m))*cos( (pi*l*(m-0.5)) /N);
+    end
+    MFCC(l) = DCTout_temp;
+end
+
+
+% figure;
+% for m=1:N
+%     
+%     plot(H(m, :), 'LineWidth', 2);
+%     hold on
+% end
+% axis tight
diff --git a/casestudy1/question1/question1.mlx b/casestudy1/question1/question1.mlx
diff --git a/casestudy1/question1/question1_fig.jpg b/casestudy1/question1/question1_fig.jpg
diff --git a/casestudy1/question1/spec_calc.m b/casestudy1/question1/spec_calc.m
@@ -0,0 +1,35 @@
+function [b,f,t] = spec_calc(data,NFFT,fs,windowSig,o_lap)
+
+L = length(windowSig);
+M = L - o_lap;
+datalen = length(data);
+
+noOfFrames = round(datalen/M);
+frameInd = 1;
+for i=1:noOfFrames
+    w_start = (i-1)*M + 1;
+    w_end = w_start + L -1;
+
+    if(w_start < 1)
+        w_start = 1;
+    end;
+    if(w_end > length(data))
+        break;
+
+    end;
+
+
+
+    h = windowSig;
+    Signal = data(w_start:w_end);
+
+    sn = Signal.*h;
+    XkTmep = fft(sn,NFFT);
+    Xk(frameInd,:) = XkTmep(1:NFFT/2);
+
+
+    frameInd = frameInd + 1;
+end
+f = (1:1:NFFT/2)*(fs/NFFT);
+t = 0:(M/fs):(datalen/fs);
+b = Xk';
diff --git a/casestudy1/question2/STEn_STM_calc.m b/casestudy1/question2/STEn_STM_calc.m
@@ -0,0 +1,26 @@
+function [STEn, STMag] = STEn_STM_calc(speech, WL_samp, WS_Samp, fs)
+
+
+
+Num_frames = round(  (length(speech)-WL_samp)/WS_Samp );
+STEn = zeros(1,Num_frames);
+STMag = zeros(1,Num_frames);
+
+for i=1:Num_frames
+    startsamp = (i-1)*WS_Samp+1;
+    EndSamp = startsamp+WL_samp-1;
+    CurrSpeech  = speech(startsamp :EndSamp);
+    CurrSpeechWin = (CurrSpeech).*hamming(WL_samp);
+    STEn(i) = sum(CurrSpeechWin.*CurrSpeechWin);
+    STMag(i) = sum(abs(CurrSpeechWin));
+
+
+%     if(i==11)
+%         
+%         FFTSpec = fft(CurrSpeechWin, 512);
+%         freq = (0:256)*(fs/512);
+%         figure; plot (freq, 20*log10( abs(FFTSpec(1:257))),'LineWidth',2);
+%         temp=1;
+%     end
+end
+
diff --git a/casestudy1/question2/abhishek_aiu.wav b/casestudy1/question2/abhishek_aiu.wav
diff --git a/casestudy1/question2/abhishek_second.wav b/casestudy1/question2/abhishek_second.wav
diff --git a/casestudy1/question2/calc_MFCC.m b/casestudy1/question2/calc_MFCC.m
@@ -0,0 +1,68 @@
+function [MFCC, logFiltBankOut, logFiltBankFreq] = calc_MFCC(speech, N_filters, N_FFT, fs)
+% calculate the MFCC now.
+fmax = fs/2;
+Mel_fmax = 1127*log(1+ (fmax/700));
+
+% obtain the center frequencies in mel scale
+N = N_filters;
+% N_FFT = 512;
+m=0:(N+1);
+c = m*(Mel_fmax/(N+1));
+
+% convert these to linear frequencies
+h = 700* (exp(c/1127) - 1);
+
+% convert to DFT indices
+g = round( (N_FFT*h)/fs) + 1; % +1 added due to MatLab indexing 1
+
+% Obtain the filterbank coefficients
+temp=1;
+
+H = zeros(N, (N_FFT/2)+1); 
+for m=2:N+1
+    for k=1:(N_FFT/2)+1
+        if( k< g(m-1) || k>g(m+1)) 
+            H(m-1, k) = 0;
+        else
+            if(k>=g(m-1) && k<=g(m))
+                H(m-1, k) = (k-g(m-1))/(g(m)-g(m-1));
+            else
+                H(m-1, k) = (g(m+1)-k)/(g(m+1)-g(m));
+            end
+
+        end
+    end
+end
+
+
+%%%%%%%%%%%%calculating MFCC %%%%%%%%%%
+
+spec = abs(  fft(speech.*hamming(length(speech)), N_FFT));
+
+% Filter the spectrum using 
+spec = spec( 1: (N_FFT/2)+1);
+
+% size(spec)
+% size(H)
+Fout = H* spec;
+logFiltBankOut = log(Fout);
+logFiltBankFreq = g(2:N+1) * (fs/N_FFT);
+
+% DCT calculation
+
+for l=1:13
+    DCTout_temp=0;
+    for m=1:N
+        DCTout_temp  = DCTout_temp+ log(Fout(m))*cos( (pi*l*(m-0.5)) /N);
+    end
+    MFCC(l) = DCTout_temp;
+end
+
+
+% figure;
+% for m=1:N
+%     
+%     plot(H(m, :), 'LineWidth', 2);
+%     hold on
+% end
+% axis tight
diff --git a/casestudy1/question2/figure1.jpg b/casestudy1/question2/figure1.jpg
diff --git a/casestudy1/question2/figure2.jpg b/casestudy1/question2/figure2.jpg
diff --git a/casestudy1/question2/figure3.jpg b/casestudy1/question2/figure3.jpg
diff --git a/casestudy1/question2/figure4.jpg b/casestudy1/question2/figure4.jpg
diff --git a/casestudy1/question2/figure5.jpg b/casestudy1/question2/figure5.jpg
diff --git a/casestudy1/question2/figure6.jpg b/casestudy1/question2/figure6.jpg
diff --git a/casestudy1/question2/figure7.jpg b/casestudy1/question2/figure7.jpg
diff --git a/casestudy1/question2/question2.mlx b/casestudy1/question2/question2.mlx
diff --git a/casestudy1/question2/spec_calc.m b/casestudy1/question2/spec_calc.m
@@ -0,0 +1,35 @@
+function [b,f,t] = spec_calc(data,NFFT,fs,windowSig,o_lap)
+
+L = length(windowSig);
+M = L - o_lap;
+datalen = length(data);
+
+noOfFrames = round(datalen/M);
+frameInd = 1;
+for i=1:noOfFrames
+    w_start = (i-1)*M + 1;
+    w_end = w_start + L -1;
+
+    if(w_start < 1)
+        w_start = 1;
+    end;
+    if(w_end > length(data))
+        break;
+
+    end;
+
+
+
+    h = windowSig;
+    Signal = data(w_start:w_end);
+
+    sn = Signal.*h;
+    XkTmep = fft(sn,NFFT);
+    Xk(frameInd,:) = XkTmep(1:NFFT/2);
+
+
+    frameInd = frameInd + 1;
+end
+f = (1:1:NFFT/2)*(fs/NFFT);
+t = 0:(M/fs):(datalen/fs);
+b = Xk';
diff --git a/casestudy1/question3/STEn_STM_calc.m b/casestudy1/question3/STEn_STM_calc.m
@@ -0,0 +1,26 @@
+function [STEn, STMag] = STEn_STM_calc(speech, WL_samp, WS_Samp, fs)
+
+
+
+Num_frames = round(  (length(speech)-WL_samp)/WS_Samp );
+STEn = zeros(1,Num_frames);
+STMag = zeros(1,Num_frames);
+
+for i=1:Num_frames
+    startsamp = (i-1)*WS_Samp+1;
+    EndSamp = startsamp+WL_samp-1;
+    CurrSpeech  = speech(startsamp :EndSamp);
+    CurrSpeechWin = (CurrSpeech).*hamming(WL_samp);
+    STEn(i) = sum(CurrSpeechWin.*CurrSpeechWin);
+    STMag(i) = sum(abs(CurrSpeechWin));
+
+
+%     if(i==11)
+%         
+%         FFTSpec = fft(CurrSpeechWin, 512);
+%         freq = (0:256)*(fs/512);
+%         figure; plot (freq, 20*log10( abs(FFTSpec(1:257))),'LineWidth',2);
+%         temp=1;
+%     end
+end
+
diff --git a/casestudy1/question3/abhishek_aiu.wav b/casestudy1/question3/abhishek_aiu.wav
diff --git a/casestudy1/question3/abhishek_second.wav b/casestudy1/question3/abhishek_second.wav
diff --git a/casestudy1/question3/calc_MFCC.m b/casestudy1/question3/calc_MFCC.m
@@ -0,0 +1,68 @@
+function [MFCC, logFiltBankOut, logFiltBankFreq] = calc_MFCC(speech, N_filters, N_FFT, fs)
+% calculate the MFCC now.
+fmax = fs/2;
+Mel_fmax = 1127*log(1+ (fmax/700));
+
+% obtain the center frequencies in mel scale
+N = N_filters;
+% N_FFT = 512;
+m=0:(N+1);
+c = m*(Mel_fmax/(N+1));
+
+% convert these to linear frequencies
+h = 700* (exp(c/1127) - 1);
+
+% convert to DFT indices
+g = round( (N_FFT*h)/fs) + 1; % +1 added due to MatLab indexing 1
+
+% Obtain the filterbank coefficients
+temp=1;
+
+H = zeros(N, (N_FFT/2)+1); 
+for m=2:N+1
+    for k=1:(N_FFT/2)+1
+        if( k< g(m-1) || k>g(m+1)) 
+            H(m-1, k) = 0;
+        else
+            if(k>=g(m-1) && k<=g(m))
+                H(m-1, k) = (k-g(m-1))/(g(m)-g(m-1));
+            else
+                H(m-1, k) = (g(m+1)-k)/(g(m+1)-g(m));
+            end
+
+        end
+    end
+end
+
+
+%%%%%%%%%%%%calculating MFCC %%%%%%%%%%
+
+spec = abs(  fft(speech.*hamming(length(speech)), N_FFT));
+
+% Filter the spectrum using 
+spec = spec( 1: (N_FFT/2)+1);
+
+% size(spec)
+% size(H)
+Fout = H* spec;
+logFiltBankOut = log(Fout);
+logFiltBankFreq = g(2:N+1) * (fs/N_FFT);
+
+% DCT calculation
+
+for l=1:13
+    DCTout_temp=0;
+    for m=1:N
+        DCTout_temp  = DCTout_temp+ log(Fout(m))*cos( (pi*l*(m-0.5)) /N);
+    end
+    MFCC(l) = DCTout_temp;
+end
+
+
+% figure;
+% for m=1:N
+%     
+%     plot(H(m, :), 'LineWidth', 2);
+%     hold on
+% end
+% axis tight
diff --git a/casestudy1/question3/fig1.jpg b/casestudy1/question3/fig1.jpg
diff --git a/casestudy1/question3/fig10.jpg b/casestudy1/question3/fig10.jpg
diff --git a/casestudy1/question3/fig2.jpg b/casestudy1/question3/fig2.jpg
diff --git a/casestudy1/question3/fig3.jpg b/casestudy1/question3/fig3.jpg
diff --git a/casestudy1/question3/fig4.jpg b/casestudy1/question3/fig4.jpg
diff --git a/casestudy1/question3/fig5.jpg b/casestudy1/question3/fig5.jpg
diff --git a/casestudy1/question3/fig6.jpg b/casestudy1/question3/fig6.jpg
diff --git a/casestudy1/question3/fig7.jpg b/casestudy1/question3/fig7.jpg
diff --git a/casestudy1/question3/fig8.jpg b/casestudy1/question3/fig8.jpg
diff --git a/casestudy1/question3/fig9.jpg b/casestudy1/question3/fig9.jpg
diff --git a/casestudy1/question3/question3.mlx b/casestudy1/question3/question3.mlx
diff --git a/casestudy1/question3/spec_calc.m b/casestudy1/question3/spec_calc.m
@@ -0,0 +1,35 @@
+function [b,f,t] = spec_calc(data,NFFT,fs,windowSig,o_lap)
+
+L = length(windowSig);
+M = L - o_lap;
+datalen = length(data);
+
+noOfFrames = round(datalen/M);
+frameInd = 1;
+for i=1:noOfFrames
+    w_start = (i-1)*M + 1;
+    w_end = w_start + L -1;
+
+    if(w_start < 1)
+        w_start = 1;
+    end;
+    if(w_end > length(data))
+        break;
+
+    end;
+
+
+
+    h = windowSig;
+    Signal = data(w_start:w_end);
+
+    sn = Signal.*h;
+    XkTmep = fft(sn,NFFT);
+    Xk(frameInd,:) = XkTmep(1:NFFT/2);
+
+
+    frameInd = frameInd + 1;
+end
+f = (1:1:NFFT/2)*(fs/NFFT);
+t = 0:(M/fs):(datalen/fs);
+b = Xk';