利用HMM的孤立字词语音识别系统

《利用HMM的孤立字词语音识别系统》由会员分享，可在线阅读，更多相关《利用HMM的孤立字词语音识别系统（12页珍藏版）》请在装配图网上搜索。

1、文档来源为 :从网络收集整理.word 版本可编辑.欢迎下载支持利用HMM的孤立字（词）语音识别系统/*/*/文件名称：recog.cpp（主程序）*/12/*/#includeemhmm.hintBR_DispMAX_ARCS;/*Displaymenttosavemultiplies(branch_prob)*/intME_DispMAX_ARCS;/*Displaymenttosavemnltiplies(mean)*/intCO_DispMAX_ARCS;/*Displaymenttosavemultiplies(covariance)*/intTrain_DispMAX_ARCS+M

2、AX_STATES;/*reallyonlyneedmax(ma,ms)*/imNum_Dimensions;intNum_Train_Sets;/*Sizeofdimensionoftrainingdata*/intTotal_Frames;intLongest_Train;/*thelengthofthelongesttrainingset*/TRAININGTrainMAX-_TRAIN_SIZE;HMMHmmMAX_REF;main(argc,argv)intargc;char*argv;inti,num_ref,result;intcntMAX_REF;stringhmm_file;

3、char*infile=NULL,*ref_list=NULL;charch_rMAX_REF8;FILE*fp;/*命令行参数设置*/if(argc5)usage(argv0);for(i=1;iargc;i+)if(argvi0!=-)fprinff(stderr,Illegalswitch%sn,argvi);usage(amv0);switch(argvi)1);caseh:ref_list=argv+i;break;casei:in_file=argv+i;break;default:fprimf(stderr,Illegalswitch%sn,argvi);usage(argv0)

4、;break;/*读取参考模型的数量，即词表的大小*/if(fp=fopen(ref_list,r)=NULL)fprinff(stderr,%s:cantopen%storead.n,ref_list);exit(0);fscanf(fp,%d,&num_ref);/*读取模型的参数*/for(i=0;iMAX_TRAIN_SIZE)fprintf(stderr,%s:toomanytestpatternsin%s.n,argv0,in_file);exit(0);/*识别*/for(i=0;iNum_Train_Sets;+i)result=dp(Train+i,numref);+cntr

5、esult;#ifTRACEprintf(%3d%3d:%sn,i+1,Traini.length,ch_rresult);fflush(stdout);#endiffree(Traini.data);printf(#识别结果#)；for(i=0;inum_ref;i+)printf(#/%4s/=%6.2f%n,ch_ri,cnti*100.0/Num_Train_Sets);/*/*函数名称：usage*/*/*函数功能：提示信息，并退出程序/*输入函数：obj-指向提示信息的指针/*/usage(obj)char*obj;fprintf(stderr,Useage:%s-hHMM-lis

6、t-iinput-filen,obj);exit(0);文件名称：(CMHMM的输入输出函数)#includecmhmm.h#includeexternintBR_Disp,ME_Disp,CO_Disp,Train_Disp;externintTotal_Frames;externintNum_Dimensions,Longest_Train;externTRAININGTrain;externHMMHmm;voidread_hmm(char*file,intrnum)FILE*fp;inttemp,i,j,k,num,mat_num,from,to,bdisp,mdisp,cdisp;fl

7、oattrans_prob;if(fp=fopen(file,r)=NULL)exit(0);if(fscanf(fp,%d,&Num_Dimensions)=EOF)exit(0);if(fscanf(fp,%d,&temp)=EOF)exit(0);Hmmrnum.num_omatrix=temp;if(fscanf(fp,%d,&temp)=EOF)exit(0);Hmmrnum.num_mixture=temp;for(i=0;iHmmrnum.num_omatrix;+i)temp=i*Hmmrnum.num_mixture;BR_Dispi=temp;temp*=Num_Dimen

8、sions;ME_Dispi=temp;temp*=Num_Dimensions;CO_Dispi=temp;temp=FLOATSIZE*Hmmrnum.num_omatrix*Hmmrnum.num_mixture;if(Hmmrnum.branch_prob=(float*)malloc(temp)=NULL)exit(0);temp*=Num_Dimensions;if(Hmmrnum.mean=(float*)malloc(temp)=NULL)exit(0);temp*=Num_Dimensions;if(Hmmrnum.covariance=(float*)malloc(temp

9、)=NULL)exit(0);for(i=0;iHmmrnum.num_omatrix;i+)bdisp=BR_Dispi;for(j=0;jHmmrnum.num_mixture;j+)if(fscanf(fp,%f,Hmmrnum.branch_prob+bdisp+j)=EOF)exit(0);for(temp=0,j=0;jHmmrnum.num_mixture;+j,temp+=Num_Dimensions)mdisp=ME_Dispi+temp;cdisp=CO_Dispi+temp*Num_Dimensions;for(k=0;kNum_Dimensions;k+)if(fsca

10、nf(fp,%f,Hmmrnum.mean+mdisp+k)=EOF)exit(0);for(k=0;kNum_Dimensions*Num_Dimensions;k+)if(fscanf(fp,%f,Hmmrnum.covariance+cdisp+k)=EOF)exit(0);if(fscanf(fp,%d,&temp)=EOF)exit(0);Hmmrnum.num_states=temp;if(Hmmrnum.states=(STATE*)malloc(sizeof(STATE)*Hmmrnum.num_states)=NULL)exit(0);for(i=0;iHmmrnum.num

11、_states;i+)Hmmrnum.statesi.label=i;Hmmrnum.statesi.num_from=Hmmrnum.statesi.num_to=0;Hmmrnum.statesi.is_initial=Hmmrnum.statesi.is_final=FALSE;if(fscanf(fp,%d,&temp)=EOF)exit(0);Hmmrnum.num_initial=temp;for(i=0;iHmmrnum.num_initial;i+)if(fscanf(fp,%d,&num)=EOF)exit(0);Hmmrnum.statesnum.is_initial=TR

12、UE;if(fscanf(fp,%d,&temp)=EOF)exit(0);Hmmrnum.num_final=temp;for(i=0;iHmmrnum.num_final;i+)if(fscanf(fp,%d,&num)=EOF)exit(0);Hmmrnum.statesnum.is_final=TRUE;if(fscanf(fp,%d,&temp)=EOF)exit(0);Hmmrnum.num_arcs=temp;if(Hmmrnum.transitions=(TRANSITION*)malloc(sizeof(TRANSITION)*Hmmrnum.num_arcs)=NULL)e

13、xit(0);for(i=0;i=Hmmrnum.num_omatrix)|(mat_numHmmrnum.num_states-1)|(fromHmmrnum.num_states-1)|(to0)fprintf(stderr,read_hmm:Illegalfrom(%d)-to(%d)n,from,to);exit(0);if(trans_prob100)exit(0);if(fs=fopen(f_list,r)=NULL)fprintf(stderr,red_data:cantopen%storeadn,f_list);exit(0);for(cnt=0;fscanf(fs,%s,f_

14、name)!=EOF;+cnt)if(cnt=MAX_TRAIN_SIZE)printf(Warninginread_data:numberofdatasets(%s)isforcedinto%d.n,f_list,MAX_TRAIN_SIZE);fflush(stdout);break;if(fp=fopen(f_name,rb)=NULL)exit(0);for(length=0;(m=fread(buff,FLOATSIZE,(Num_Dimensions+1),fp)=(Num_Dimensions+1);+length);if(Traincnt.data=(float*)malloc

15、(FLOATSIZE*Num_Dimensions*length)=NULL)fprintf(stderr,read_data:notenoughspacefor%sn,f_list);exit(EOF);rewind(fp);for(i=j=0;(m=fread(buff,FLOATSIZE,(Num_Dimensions+1),fp)=(Num_Dimensions+1);+i)for(k=1;kmax)max=length;fclose(fs);Total_Frames=total_length;Longest_Train=max;for(i=j=0;jlength;data=train

16、-data;/*memoryallocationforwork*/if(pn=(double*)malloc(Longest_Train*MAX_STATES*DOUBLEESIZE)=NULL)exit(0);if(outprob_buff=(double*)malloc(Longest_Train)*MAX_ARCS*DOUBLESIZE)=NULL)exit(0);for(s=i=0;sMAX_STATES;i+=Longest_Train)Outprob_Disps+=i;for(n=0;nnum_ref;n+)/*step1*/s=MAX_STATES*Longest_Train;f

17、or(i=0;is;i+)pni=INFINT;/*clearpn*/for(s=0;sHummn.num_omatrix;+s)if(Hmmn.statess.is_initial)pnTrain_Disps=0.0;for(i=0;iMAX_ARCA*Longest_Train;outprob_buffi+=INFINIT);for(s=0;sHummn.num_omatrix;s+)for(i=0;ilength;i+)sum_prob=0.0;for(j=0;jHummn.num_mixture;j+)branch_prob=Hummn.branch_probBR_disps+j;if

18、(branch_prob=0.0)continue;sum_prob+=pow(branch_prob,5.0)*cal_outprob(&(Hummn),n,s,i,j);outprob_buffOutprob_Disps+i=log(sum_prob);for(i=0;ilength;i+)/*step2*/for(s=0;sHummn.num_states;s+)/*step3*/index=Train_Disps+i;stop=&(Hummn.statess.trans_toHummn.statess.num_to);for(cptr=&(Hummn.statess.trans_to0

19、);cptrout_prob_index=NULL_TRANSITION)prev_pn=pn+(Train_Disptptr-origin+i);if(*prev_pnINFINIT)px=*prev_pn+tptr-trans_prob;if(pxpnindex)pnindex=px;elseif(i0)prev_pn=pn+(Train_Disaptptr-origin+i-1);if(*prev_pnINFINIE)out_prob=*(outprob_buff+Outprob_Disptptr-out_prob_index+i);if(out_probINFINIT)px=*prev

20、_pn+(tptr-trans_prob)+out_prob;if(pxpnindex)pnindex=px;pn=pnTrain_Disp(Hummn.num_states)-1+length-1;free(pn);free(outprob_buff);nhat=0;for(n=1;npnhat)nhat=n;returnnhat;/*/*函数名称：cal-outprob*/*/*/*函数功能：计算特征向量序列输入到HUMM室的输出概率的大小/*输入参数：humm-HUMM的参数；ref-num-HUMM的序号；onum-具有输出字符的/*的状态的序号；inum-输入特征向量的个数；mnum

21、-混合概率密度的序号*/*返回值：HMM的输出概率*/*/doublecal_outprob(HUMM*hmm,intref_num,intonum,intinum,intmnum)doublea100,b,out_prob;inti,j,mdisp,cdsp,ddisp;if(Num_Dismensions100)exit(0);i=mnum;ddisp=BR_Disaponum+i;i*=Num_Dimensions;mdisp=ME_Disponum+i;i*=Num_Dimensions;cdisp=CO_Disponum+i;out_prob=0.0;for(i=0;imeanmdi

22、sp+i;for(i=0;i=Num_Dimensions;i+)b=0.0;for(i=0;jMIN_OUTPROB)?out_prob;MIN_OUTPROB;文件名称：cmhmm.h#includecmhmm.h#defineM_PI3.846externintNum_Dimensions;externHMMHmm;externintCO_Disp,BR_Disp;double*Inv_CovarianceMAX_REF;double*Det_CovarianceMAX_REF;voidcal_invdet(intrnum)inti,j,k,l,m;intwork100;intlr,iw

23、,cdisp,ddisp;doubleeps=1.0e-20,w,wmax,pivot,api;if(Num_Dimensions100)exit(0);/*memoryallocation*/i=Hmmrnum.num_omatrix*Hmmrnum.num_mixture;if(Det_Covariancernum=(double*)malloc(i*DOUBLESIZE)=NULL)exit(0);for(m=0;mHmmrnum.num_omatrix;m+)for(l=0;lHmmrnum.num_mixture;l+)cdisp=CO_Dispm+Num_Dimensions*Nu

24、m_Dimensions*l;ddisp=BR_Dispm+1;for(i=0;iNum_Dimensions;i+)for(j=0;jNum_Dimensions;j+)Inv_Covariancernumcdisp+i*Num_Dimensions+j=Hmmrnum.covariancecdisp+i*Num_Dimensions+j;Det_Covariancernumddisp=1.0;for(i=0;iNum_Dimensions;i+)worki=i;for(k=0;kNum_Dimensions;k+)wmax=0.0;for(i=k;iNum_Dimensions;i+)w=

25、fabs(Inv_Covariancernumcdisp+k*Num_Dimensions+i);if(wwmax)continue;wmax=w;lr=i;pivot=Inv_Covariancernumcdisp+k*Num_Dimensions+i;api=fabs(pivot);if(api=eps)exit(0);Det_Covariancernumddisp*=pivot;if(Det_Covariancernumddisp=0.0)printf(cal_invdet:Det_Covariance%d%disunderflow.n,rnum,ddisp);exit(0);if(lr

26、!=k)Det_Covariancernumddisp*=-1.0;iw=workk;workk=workk;workk=iw;for(j=0;jNum_Dimensions;j+)w=Inv_Covariancernumcdisp+j*Num_Dimensions+k;Inv_Covariancernumcdisp+j*Num_Dimensions+k=Inv_Covariancernumcdisp+j*Num_Dimensions+lr;w=Inv_Covariancernumcdisp+j*Num_Dimensions+lr=w;for(i=0;iNum_Dimensions;i+)In

27、v_Covariancernumcdisp+i*Num_Dimensions+k/=pivot;for(i=0;iNum_Dimensions;i+)if(i=k)continue;w=Inv_Covariancernumcdisp+k*Num_Dimensions+i;if(w=0.0)continue;for(j=0;jNum_Dimensions;j+)if(j=k)continue;Inv_Covariancernumcdisp+j*Num_Dimensions+i-=w*Inv_Covariancernumcdisp+j*Num_Dimensions+k;Inv_Covariance

28、rnumcdisp+k*Num_Dimensions+i=-w/pivot;Inv_Covariancernumcdisp+k*Num_Dimensions+k=1.0/pivot;for(i=0;iNum_Dimensions;i+)while(1)k=worki;if(k=i)break;iw=workk;workk=worki;worki=iw;for(j=0;jNum_Dimensions;j+)w=Inv_Covariancernumcdisp+i*Num_Dimensions+j;Inv_Covariancernumcdisp+i*Num_Dimensions+j=Inv_Covariancernumcdisp+k*Num_Dimensions+j;Inv_Covariancernumcdisp+k*Num_Dimensions+j=w;if(w=Det_Covariancernumddisp)0.0)fprintf(stderr,ERROR:Det_Covariance%d%d1)*sqrt(w);