我想实现一个系统来识别0-9之间的数字。我有一个TemplateSet和一个记录音频信号的测试集,数字在0-9之间。我必须在所有音频文件之间进行交叉关联,以找到它们之间的最佳相似性。之后,我应该为每个信号进行时间移位以使其同步,以便将其与其他信号进行比较以获得最佳结果。我不知道如何实现时移。
MaxSize=[0,0];
for i=1:9
TempSize=size(temp{i});
TestSize=size(test{i});
if(TempSize(1)>MaxSize(1))
MaxSize(1)=TempSize(1);
end
if(TestSize(1)>MaxSize(1))
MaxSize(1)=TestSize(1);
end
end
for j=1:9
TsSize=(MaxSize-size( temp{j}));
TpSize=(MaxSize-size( test{j}));
temp{j}(end+TsSize)=0;
test{j}(end+TpSize)=0;
end
%------------ Euclidean distance-----
sum=0;
index=zeros(1,9);
Distance=zeros(9,9);
for m =1:9
MinDistance=10000;
for n=1:9
Distance(m,n)= norm(test{m}-temp{n},2);
if (Distance< MinDistance)
index(m,n)=n;
MinDistance=Distance;
end
end
if(index(m)==m)
sum=sum+1;
end
end
%--------------------------------------------------------------
TimeDiff=zeros(9,9);
for l=1:9
for k=1:9
[c,lag]=xcorr(test{l},temp{k});
TimeDiff(l,k)=max(abs(c));
end
end
答案 0 :(得分:0)
%euclidean distance
index=zeros(1,10);
Distance=zeros(10,10);
for m =1:10
MinDistance=10000;
for n=1:10
Distance(m,n)= norm(test{m}-temp{n},2);
if (Distance < MinDistance)
index(m,n)=n;
MinDistance=Distance;
end
end
end
TimeDiff=zeros(10,10);
euclidean =zeros(10,10);
for l=1:10
for i=1:10
[c,lag]=xcorr(test{l},temp{i});
[~,I]= max(abs(c));
shifted= circshift(test{1},I);
euclidean(l,k)=norm(temp{l}-shifted,2);
end
end
%.................................................
Index=0;
for i=1:10
Min=1000000;
for j=1:10
if( euclidean(i,j)<Min )
Min=euclidean(i,j);
index=j;
end
end
end
然后交叉关联信号。