function????value=GMM（centroidcovk_allv）
mindist=0.01;%設置最小的矢量距離
L=30.001;%30times?iterate?with?average?variation?of?0.001?for?each?iteration.
em_count=0;
[dimenframe_all]=size（v）;
cov_exp=zeros（dimen1620）;
centroid_exp=zeros（dimen1620）;
%???mix（v165）;
th=（L-floor（L））*frame_all;%0.001*frame_all
finalpro=0;
for?ii=1:L%em過程L=30次
????%?????divide設置初始點時候只需要根據每個類別聚類就可以了，而下面將進行的高斯模型em過程將會在每個中心?點都與每個點關聯?取得p和b的值
????em_count=em_count+1;
????p=zeros（16frame_all）;
????b=zeros（16frame_all）;
????k_all=k_all/frame_all;%i類模型出現的概率
????for?i=1:16
????????for?j=1:frame_all
????????????if?sum（abs（centroid（:i）））~=0
????????????b（ij）=（abs（prod（cov（:i））^（-1/2）））*exp（sum（（v（:j）-centroid（:i））.^2./cov（:i））*（-1/2））;
????????????else
????????????????b（ij）=0;
????????????end
????????end
????end
????for?i=1:frame_all
????????b_sum=0;
????????for?j=1:16
????????????b_sum=b_sum+b（ji）*k_all（j）;
????????end
????????for?j=1:16
????????????if?b_sum~=0%防止沒有數據的出現
????????????????p（ji）=k_all（j）*b（ji）/b_sum;
????????????else
????????????????p（ji）=0;
????????????end
????????end
????end
????centroid=zeros（dimen16）;
????cov=zeros（di