-
Notifications
You must be signed in to change notification settings - Fork 0
/
transit_validation2.m
252 lines (206 loc) · 8.65 KB
/
transit_validation2.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
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
% transit_validation2.m
% Mapping to Y1
%
% X1 -> Y1 vs X1 -> Y2 -> Y1
% X2 -> Y1 vs X2 -> Y2 -> Y1
% X3 -> Y1 vs X3 -> Y2 -> Y1
% ...
% X2263 -> Y1 vs X2263 -> Y2 -> Y1
%
% Mapping to Y2
%
% X1 -> Y2 vs X1 -> Y1 -> Y2
% X2 -> Y2 vs X2 -> Y1 -> Y2
% X3 -> Y2 vs X3 -> Y1 -> Y2
% ...
% X2263 -> Y2 vs X2263 -> Y1 -> Y2
clear all;
load FileWav16kHz
Fs = 16000;
% maproot = '.\MappingTable\GT100old';
maproot = '.\MappingTable\GT100';
% maproot = '.\MappingTable\MT';
Y1.Fname = 'english]english239.male.N_english.R_usa.Y18.A19'; Y1.Samples = [7352, 355332]; % Sample numbers of starting and ending the waveform
Y2.Fname = 'english]english10.female.N_english.R_usa.Y35.A35'; Y2.Samples = [6928, 328688];
% Y1.Samples = [4000, 360000]; Y2.Samples = [4000, 340000];
%
% Y1.Fname = 'english]english451.male.N_english.R_usa.Y44.A44'; Y1.Samples = [5000, 415711];
% Y2.Fname = 'japanese]japanese4.male.N_japanese.R_usa.Y1.A20'; Y2.Samples = [5000, 629372];
% Y1.Fname = 'english]english165.female.N_english.R_usa.Y43.A43'; Y1.Samples = [5500, 366808];
% Y2.Fname = 'japanese]japanese26.female.N_japanese.R_usa.Y6.A44'; Y2.Samples = [5000, 510711];
% Y1.Fname = 'english]english165.female.N_english.R_usa.Y43.A43'; Y1.Samples = [5500, 366808];
% Y2.Fname = 'japanese]japanese13.male.N_japanese.R_usa.Y0.A28'; Y2.Samples = [13400, 465000];
Y1.wav = wavread(fullfile('wav16kHz', [Y1.Fname, '.wav']));
Y2.wav = wavread(fullfile('wav16kHz', [Y2.Fname, '.wav']));
V1 = load(fullfile('wav16kHzTextGridMat', [Y1.Fname, '.mat']));
V2 = load(fullfile('wav16kHzTextGridMat', [Y2.Fname, '.mat']));
% V2 = load('english10');
NX = length(accnames);
Config = {
'N1W180R0S0EdN0'
'N1W180R0S0EdN1'
'N3W180R0S0EdN1'
'N4W180R0S0EdN1'
'N5W180R0S0EdN1'
};
iConfig = 3;
%% Short pause remove or not
RemoveSP = 0; % 1 - remove Short Pause -> the misalignment of short pauses
jtmp = find(strcmp(V2.PhnStr(:, 1), 'sp'));
jtmp = V2.PhnTime([jtmp(1:end-1),jtmp(1:end-1)+1])/Fs;
%% Loop begins
for ix = 1:NX
if ix==2034
disp('dfd');
end
if strcmp(Y1.Fname, accnames{ix}) | strcmp(Y2.Fname, accnames{ix})
continue;
end
XY1.fname = fullfile(Y1.Fname, [accnames{ix}, '.mat']);
XY2.fname = fullfile(Y2.Fname, [accnames{ix}, '.mat']);
Y1Y2.fname = fullfile(Y2.Fname, [Y1.Fname, '.mat']);
Y2Y1.fname = fullfile(Y1.Fname, [Y2.Fname, '.mat']);
%% Load mapping tables of XY1, XY2, Y1Y2, Y2Y1
tmp1 = load(fullfile(maproot, Config{iConfig}, XY1.fname));
XY1.mt = tmp1.MT;
% XY1.mt = tmp1.SyncOut.Log.MappingTable;
tmp2 = load(fullfile(maproot, Config{iConfig}, XY2.fname));
XY2.mt = tmp2.MT;
% XY2.mt = tmp2.SyncOut.Log.MappingTable;
tmp3 = load(fullfile(maproot, Config{iConfig}, Y1Y2.fname));
Y1Y2.mt = tmp3.MT;
% Y1Y2.mt = tmp3.SyncOut.Log.MappingTable;
tmp4 = load(fullfile(maproot, Config{iConfig}, Y2Y1.fname));
Y2Y1.mt = tmp4.MT;
% Y2Y1.mt = tmp4.SyncOut.Log.MappingTable;
%% Normalize the duration of every X mapping to a fixed Y
% Silence at the beginning and at the end is removed and convert it to
% in second
XY1.MT = XY1.mt(find(XY1.mt(:, 2) >= Y1.Samples(1) & XY1.mt(:, 2) <= Y1.Samples(2)), :)/Fs;
XY2.MT = XY2.mt(find(XY2.mt(:, 2) >= Y2.Samples(1) & XY2.mt(:, 2) <= Y2.Samples(2)), :)/Fs;
Y1Y2.MT = Y1Y2.mt(find(Y1Y2.mt(:, 2) >= Y2.Samples(1) & Y1Y2.mt(:, 2) <= Y2.Samples(2)), :)/Fs;
Y2Y1.MT = Y2Y1.mt(find(Y2Y1.mt(:, 2) >= Y1.Samples(1) & Y2Y1.mt(:, 2) <= Y1.Samples(2)), :)/Fs;
% Polynomial fitting
% P = polyfit(XY1.MT(:, 1), XY1.MT(:, 2), 1);
% XY1.MT(:, 1) = polyval(P, XY1.MT(:, 1));
% P = polyfit(XY2.MT(:, 1), XY2.MT(:, 2), 1);
% XY2.MT(:, 1) = polyval(P, XY2.MT(:, 1));
% P = polyfit(Y1Y2.MT(:, 1), Y1Y2.MT(:, 2), 1);
% Y1Y2.MT(:, 1) = polyval(P, Y1Y2.MT(:, 1));
% P = polyfit(Y2Y1.MT(:, 1), Y2Y1.MT(:, 2), 1);
% Y2Y1.MT(:, 1) = polyval(P, Y2Y1.MT(:, 1));
% Naive fitting
C = XY1.MT(1, 1);
XY1.MT(:, 1) = (XY1.MT(:, 1) - C)/(XY1.MT(end, 1) - C)*(XY1.MT(end, 2) - C) + C;
C = XY2.MT(1, 1);
XY2.MT(:, 1) = (XY2.MT(:, 1) - C)/(XY2.MT(end, 1) - C)*(XY2.MT(end, 2) - C) + C;
C = Y1Y2.MT(1, 1);
Y1Y2.MT(:, 1) = (Y1Y2.MT(:, 1) - C)/(Y1Y2.MT(end, 1) - C)*(Y1Y2.MT(end, 2) - C) + C;
C = Y2Y1.MT(1, 1);
Y2Y1.MT(:, 1) = (Y2Y1.MT(:, 1) - C)/(Y2Y1.MT(end, 1) - C)*(Y2Y1.MT(end, 2) - C) + C;
%% Checking the normalization effect by plotting
% figure(1);
% plot(XY1.mt(:, 1),XY1.mt(:, 2)); hold on;
% plot(Y2Y1.mt(:, 1), Y2Y1.mt(:, 2),'r'); hold off;
% title('Before Normaliztion');
%
% figure(2);
% plot(XY1.MT(:, 1),XY1.MT(:, 2)); hold on;
% plot(Y2Y1.MT(:, 1), Y2Y1.MT(:, 2),'r'); hold off;
% title('After Normaliztion');
%% create Composite mapping XY1Y2, XY2Y1
Y1.Grid = XY1.MT(:, 2);
Y2.Grid = XY2.MT(:, 2);
XY1Y2.MT = compositmap(XY1.MT, Y1Y2.MT, Y2.Grid);
XY2Y1.MT = compositmap(XY2.MT, Y2Y1.MT, Y1.Grid);
% Mean subtraction
XY1.MT = bsxfun(@minus, XY1.MT, mean(XY1.MT));
XY2Y1.MT = bsxfun(@minus, XY2Y1.MT, mean(XY2Y1.MT));
XY2.MT = bsxfun(@minus, XY2.MT, mean(XY2.MT));
XY1Y2.MT = bsxfun(@minus, XY1Y2.MT, mean(XY1Y2.MT));
% Variance Normalization
% XY1.MT = bsxfun(@times, XY1.MT, 1./std(XY1.MT));
% XY2Y1.MT = bsxfun(@times, XY2Y1.MT, 1./std(XY2Y1.MT));
% XY2.MT = bsxfun(@times, XY2.MT, 1./std(XY2.MT));
% XY1Y2.MT = bsxfun(@times, XY1Y2.MT, 1./std(XY1Y2.MT));
%% Compare the direct vs indirect map
% figure(1);
% plot(XY2Y1.MT(:, 1), XY2Y1.MT(:, 2), '.-'); hold on;
% plot(XY1.MT(:, 1), XY1.MT(:, 2), '.-r');
% % plot(Y1.wav+10, (1:length(Y1.wav))/Fs); hold off; grid on;
%
%
% figure(2);
% plot(XY1Y2.MT(:, 1), XY1Y2.MT(:, 2), '.-'); hold on;
% plot(XY2.MT(:, 1), XY2.MT(:, 2), '.-r');
% plot(Y2.wav+10, (1:length(Y2.wav))/Fs); hold off; grid on;
%% Remove SP (SHort Pause) ?
if RemoveSP
i_SP = find( Y2.Grid > jtmp(1, 1) & Y2.Grid <= jtmp(1, 2));
for i = 2:size(jtmp, 1)-1
i_SP = [i_SP; find( Y2.Grid > jtmp(i, 1) & Y2.Grid <= jtmp(i, 2))];
end
i_SP = [i_SP; find(Y2.Grid > jtmp(end, 1))];
XY2.MT(i_SP, :) = [];
XY1Y2.MT(i_SP, :) = [];
end
%% Error calculation
Diff.XY1 = XY1.MT(:, 1) - XY2Y1.MT(:, 1); N1 = size(Diff.XY1, 1);
Diff.XY2 = XY2.MT(:, 1) - XY1Y2.MT(:, 1); N2 = size(Diff.XY2, 1);
% Diff.XY1 = Diff.XY1 - mean(Diff.XY1);
% Diff.XY2 = Diff.XY2 - mean(Diff.XY2);
Err(ix).XY1 = sum(abs(Diff.XY1))/N1;
Err(ix).XY2 = sum(abs(Diff.XY2))/N2;
PC20(ix).XY1 = sum(abs(Diff.XY1) < .02)/N1;
PC20(ix).XY2 = sum(abs(Diff.XY2) < .02)/N2;
PC25(ix).XY1 = sum(abs(Diff.XY1) < .025)/N1;
PC25(ix).XY2 = sum(abs(Diff.XY2) < .025)/N2;
% Err(ix).XY1, Err(ix).XY2
%% status printing
if mod(ix, 100) == 0
fprintf('[ %d / %d]\n', ix, NX);
end
end
%% Pick speakers with least transitive validation error
SpkrSubset(1, 1).Fname = Y1.Fname;
SpkrSubset(1, 1).Kind = '80% least transitive error';
[yy, ii] = sort([Err.XY1]);
SpkrSubset(1, 1).bool = zeros(1, NX);
SpkrSubset(1, 1).bool(ii(1:round(NX*.8))) = 1;
SpkrSubset(1, 2).Fname = Y1.Fname;
SpkrSubset(1, 2).Kind = '60% least transitive error';
[yy, ii] = sort([Err.XY1]);
SpkrSubset(1, 2).bool = zeros(1, NX);
SpkrSubset(1, 2).bool(ii(1:round(NX*.6))) = 1;
SpkrSubset(1, 3).Fname = Y1.Fname;
SpkrSubset(1, 3).Kind = '60% Random pick';
SpkrSubset(1, 3).bool = zeros(1, NX);
SpkrSubset(1, 3).bool(randsample(NX, round(.6*NX))) = 1;
SpkrSubset(2, 1).Fname = Y2.Fname;
SpkrSubset(2, 1).Kind = '80% least transitive error';
[yy, ii] = sort([Err.XY2]);
SpkrSubset(2, 1).bool = zeros(1, NX);
SpkrSubset(2, 1).bool(ii(1:round(NX*.8))) = 1;
SpkrSubset(2, 2).Fname = Y2.Fname;
SpkrSubset(2, 2).Kind = '60% least transitive error';
[yy, ii] = sort([Err.XY2]);
SpkrSubset(2, 2).bool = zeros(1, NX);
SpkrSubset(2, 2).bool(ii(1:round(NX*.6))) = 1;
SpkrSubset(2, 3).Fname = Y2.Fname;
SpkrSubset(2, 3).Kind = '60% Random pick';
SpkrSubset(2, 3).bool = zeros(1, NX);
SpkrSubset(2, 3).bool(randsample(NX, round(.6*NX))) = 1;
save SpkrSubset SpkrSubset
%% Error
% .1838, .2233 --> polyfit
% .1597, .1891 --> Naive Fit (SP removed -- .1902)
% .1734, .1992 --> N1(Naive)
% .1983, .2264 --> N1(Polyfit)
median([Err.XY1])
median([Err.XY2])
figure(5);
plot([Err.XY1]); xlabel('# Speaker'); ylabel('Error in [Sec]');
title(Y1.Fname, 'Interpreter', 'None');
figure(6);
plot([Err.XY2]); xlabel('# Speaker'); ylabel('Error in [Sec]');
title(Y2.Fname, 'Interpreter', 'None');