melp_msvq.asv

语音编码中的MELP的经典估计

%Quantization of Prediction Coefficients
%Input:
%        lpcs(LPC parameters)
%        f(corresponding LSF parameters)
%Output:
%        g(Index)
function g=melp_msvq(lpcs,f)
global stage1;
global stage2;
for j=1:10													%Get the Weighs
   w(j)=1+exp(-i*f(j)*(1:10))*lpcs';
end
w=abs(w);
w=w.^0.3;
w(9)=w(9)*0.64;
w(10)=w(10)*0.16;

%d(m,1) is the judgement.
%d(m,2:11) is the minus of the vector and the codeword
%d(m,12:15) is the codeword.
d(1:8,1:12)=10000000;
for s=1:128
   delta=f-stage1((s-1)*10+1:s*10);
   temp=w*(delta.^2)';
   m=1;
   while m<9
      if temp<d(m,1)
         d(m+1:9,:)=d(m:8,:);
         d(m,1)=temp;
         d(m,2:11)=delta;
         d(m,12)=s;
         break;
      end
      m=m+1;
   end
end
for i=1:3
   e=d;
   d(1,2:11)=e(1,2:11)-stage2(i,1:10);
   d(1,1)=w*(d(1,2:11).^2)';
   d(1,12:12+i)=[e(1,12:11+i),1];
   for m=2:8
      delta=e(m,2:11)-stage2(1,1:10);
      temp=w*(delta.^2)';
      for num=1:(m-1)
         if temp<d(num,1)
         	d(num+1:9,:)=d(num:8,:);
         	d(num,1)=temp;
         	d(num,2:11)=delta;
         	d(num,12:11+i)=e(1,12:11+i);
         	d(num,12+i)=1;
         	break;
         end
   	end
      if temp>=d(m-1,1)
         d(m,2:11)=delta;
         d(m,1)=temp;
         d(m,12:12+i)=[e(1,12:11+i),1];
      end
   end
   for j=1:8
      for k=2:64
         delta=e(j,2:11)-stage2(i,(k-1)*10+1:k*10);
         temp=w*(delta.^2)';
         for n=1:8
            if temp<d(n,1)
               d(n+1:9,:)=d(n:8,:);
               d(n,1)=temp;
               d(n,2:11)=delta;
               d(n,12:11+i)=e(j,12:11+i);
               d(n,12+i)=k;
               break;
            end
         end
      end
   end
end
g=d(1,12:15);