-
Notifications
You must be signed in to change notification settings - Fork 0
/
pitch_shift_demo.m
157 lines (134 loc) · 6.64 KB
/
pitch_shift_demo.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
suspect1 = 'C:/Users/SiddGururani/Desktop/Stevie wonder samples/Inorganic Dataset/pitch_shifted/01.wav';
sample = 'C:/Users/SiddGururani/Desktop/Stevie wonder samples/Inorganic Dataset/New_sample/01.wav';
suspect2 = 'C:/Users/SiddGururani/Desktop/Stevie wonder samples/Inorganic Dataset/New_songs/02.wav';
[data_orig,fs] = audioread(sample);
[data_copy1,fs] = audioread(suspect1);
% [data_copy2,fs] = audioread(suspect2);
% data_orig = bsxfun(@rdivide, data_orig, rms(data_orig,1));
% data_copy = bsxfun(@rdivide, data_copy, rms(data_copy,1));
data_orig = downsample(data_orig,2);
data_copy1 = downsample(data_copy1,2);
% data_copy2 = downsample(data_copy2,2);
fs = fs/2;
%% Computing the STFT spectrograms
data_orig = mean(data_orig,2);
data_copy1 = mean(data_copy1,2);
% data_copy2 = mean(data_copy2,2);
window = 4096;
hop = 1024;
Xo = spectrogram(data_orig, window, window-hop);
Xs = spectrogram(data_copy1, window, window-hop);
% Xs2 = spectrogram(data_copy2, window, window-hop);
% f = 0:(length(freq)-1);
% f = f*((fs/2)/length(freq));
%
% midi = 69 + 12*log2(f/440);
%% Performing Non-negative Matrix Factorization on the original sample spectrogram
n = 0; % no pitch shifts
k = 6;
[Bo, Ho] = nnmf(abs(Xo), k);
% Check for low-rank < k
rank_check = sum(Ho,2);
if ~isempty(find(rank_check == 0))
Ho(find(rank_check == 0),:) = [];
Bo(:,find(rank_check == 0)) = [];
k = k - numel(find(rank_check == 0));
end
% figure; imagesc([1:size(Ho,2)].*hop/fs,[2048:0],(abs(Xo))); xlabel('Time (s)'); ylabel('Frequency Bin');title('Magnitude Spectrogram of Sample');
%
% hFig = figure;
% set(hFig, 'Position', [450 100 1000 800])
% subplot(2,3,1); plot(Bo(:,1)); xlabel('Frequency Bin'); ylabel('Magnitude'); title('First Template'); xlim([0, 2049]);
% subplot(2,3,2); plot(Bo(:,2)); xlabel('Frequency Bin'); ylabel('Magnitude'); title('Second Template'); xlim([0, 2049]);
% subplot(2,3,3); plot(Bo(:,3)); xlabel('Frequency Bin'); ylabel('Magnitude'); title('Third Template'); xlim([0, 2049]);
% subplot(2,3,4); plot(Bo(:,4)); xlabel('Frequency Bin'); ylabel('Magnitude'); title('Fourth Template'); xlim([0, 2049]);
% subplot(2,3,5); plot(Bo(:,5)); xlabel('Frequency Bin'); ylabel('Magnitude'); title('Fifth Template'); xlim([0, 2049]);
% subplot(2,3,6); plot(Bo(:,6)); xlabel('Frequency Bin'); ylabel('Magnitude'); title('First Template'); xlim([0, 2049]);
% suptitle('Factorized Sample Spectrogram to Get Templates');
%
% hFig = figure;
% set(hFig, 'Position', [450 100 1000 800])
% subplot(2,3,1); plot([1:size(Ho,2)].*hop/fs,Ho(1,:)); xlabel('Time (s)'), ylabel('Magnitude'); title('First Activation');
% subplot(2,3,2); plot([1:size(Ho,2)].*hop/fs,Ho(2,:)); xlabel('Time (s)'), ylabel('Magnitude'); title('Second Activation');
% subplot(2,3,3); plot([1:size(Ho,2)].*hop/fs,Ho(3,:)); xlabel('Time (s)'), ylabel('Magnitude'); title('Third Activation');
% subplot(2,3,4); plot([1:size(Ho,2)].*hop/fs,Ho(4,:)); xlabel('Time (s)'), ylabel('Magnitude'); title('Fourth Activation');
% subplot(2,3,5); plot([1:size(Ho,2)].*hop/fs,Ho(5,:)); xlabel('Time (s)'), ylabel('Magnitude'); title('Fifth Activation');
% subplot(2,3,6); plot([1:size(Ho,2)].*hop/fs,Ho(6,:)); xlabel('Time (s)'), ylabel('Magnitude'); title('Sixth Activation');
% suptitle('Factorized Sample Spectrogram to Get Activations');
%
% figure; imagesc([1:size(Ho,2)].*hop/fs,[0:2048],(Bo*Ho)); xlabel('Time (s)'); ylabel('Frequency Bin'); title('Reconstructed Magnitude Spectrogram of Sample');
%
% fprintf('Press a key to continue...\n');
% pause;
%% Computing pitch shifted templates for detecting pitch shifted samples
N = numel(Bo(:,1));
Bo_concat = Bo;
% n : number of semi-tone shifts.
n = 24;
% shift down
for i=1:12
t1 = 1:N;
t2 = (t1)/(2^(-i/12));
B_shift = interp1(t1, Bo, t2,'linear');
if t1(end)<t2(end)
B_shift(floor(t1(end)/t2(end)*t1(end)):end,:) = 0;
end
Bo_concat = [Bo_concat B_shift];
end
%shift up
for i=1:12
t1 = 1:N;
t2 = (t1)/(2^(i/12));
B_shift = interp1(t1, Bo, t2,'linear');
if t1(end)<t2(end)
B_shift(floor(t1(end)/t2(end)*t1(end)):end,:) = 0;
end
Bo_concat = [Bo_concat B_shift];
end
clear B_shift Bo
%% Performing partially fixed NMF using the precomputed template matrix
% [Bo1, Ho_hypo1, ~, ~, err] = PfNmf(abs(Xs1), Bo, [], [], [], 0, 0);
% [Bo1, Ho_hypo2, ~, ~, err] = PfNmf(abs(Xs2), Bo, [], [], [], 0, 0);
% figure; imagesc([1:size(Ho_hypo1,2)].*hop/fs,[0:2048],(Bo*Ho_hypo1)); xlabel('Time (s)'); ylabel('Frequency Bin'); title('Reconstructed Spectrogram of Sample using Song Activations: Match');
% figure; imagesc([1:size(Ho_hypo2,2)].*hop/fs,[0:2048],(Bo*Ho_hypo2)); xlabel('Time (s)'); ylabel('Frequency Bin'); title('Reconstructed Spectrogram of Sample using Song Activations: No Match');
% fprintf('Press a key to continue...\n');
% pause;
%% Performing partially fixed NMF using the pitch-shift templates concatenated as well.
% use either one of the PFNMF sections
[Bo1, Ho_hypo, ~, ~, err] = PfNmf(abs(Xs), Bo_concat, [], [], [], 0, 0);
%% Normalize the two activation matrices before computing correlation
% DO NOT RUN
% for i = 1:k
% Ho(i,:) = Ho(i,:)/norm(Ho(i,:),1);
% end
%
% for i = 1:(n+1)*k
% Ho_hypo(i,:) = Ho_hypo(i,:)/norm(Ho_hypo(i,:),1);
% end
%% Computing correlation and subsequently the occurences between activation
% matrices of original sample and suspected copy
% [corrMat, instants] = FastCorrelate(Ho_hypo, Ho);
% corrMat(corrMat<0) = 0;
% [corr1, lags1] = corr_activations(Ho,Ho_hypo1);
% figure; plot(lags1*1024/22050,corr1); xlabel('Time (s)'); ylabel('Correlation'); title('6 Correlation Functions Between Corresponding Activations: Match');
%
% [corr2, lags2] = corr_activations(Ho,Ho_hypo2);
% figure; plot(lags2*1024/22050,corr2); xlabel('Time (s)'); ylabel('Correlation'); title('6 Correlation Functions Between Corresponding Activations: No Match');
%
% fprintf('Press a key to continue...\n');
% pause;
%% Peak picking. Currently very basic hard threshold. Add better post-processing.
i = 1;
[corr, lags] = corr_activations(Ho,Ho_hypo((i-1)*k+1:i*k,:));
prod_corr = prod(corr).^(1/k);
[peaks, loc] = findpeaks(abs(prod_corr));
loc(peaks<0.6) = [];
peaks(peaks<0.6) = [];
figure; plot(lags*1024/22050,prod_corr); xlabel('Time (s)'); ylabel('Correlation'); title('Geometric Mean of Correlations for all Activations: Sample Match, Incorrect Pitch');
i = 17;
[corr, lags] = corr_activations(Ho,Ho_hypo((i-1)*k+1:i*k,:));
prod_corr = prod(corr).^(1/k);
[peaks, loc] = findpeaks(abs(prod_corr));
loc(peaks<0.6) = [];
peaks(peaks<0.6) = [];
figure; plot(lags*1024/22050,prod_corr); xlabel('Time (s)'); ylabel('Correlation'); title(['Geometric Mean of Correlations for all Activations: Sample Match, Correct Pitch ' num2str(mod(i-1,12)) 'Higher']);