hquant.c

该压缩包为最新版htk的源代码,htk是现在比较流行的语音处理软件,请有兴趣的朋友下载使用

/* ----------------------------------------------------------- */
/*                                                             */
/*                          ___                                */
/*                       |_| | |_/   SPEECH                    */
/*                       | | | | \   RECOGNITION               */
/*                       =========   SOFTWARE                  */ 
/*                                                             */
/*                                                             */
/* ----------------------------------------------------------- */
/*         Copyright: Microsoft Corporation                    */
/*          1995-2000 Redmond, Washington USA                  */
/*                    http://www.microsoft.com                 */
/*                                                             */
/*   Use of this software is governed by a License Agreement   */
/*    ** See the file License for the Conditions of Use  **    */
/*    **     This banner notice must not be removed      **    */
/*                                                             */
/* ----------------------------------------------------------- */
/*       File: HQuant.c: VQ table generation                   */
/* ----------------------------------------------------------- */

char *hquant_version = "!HVER!HQuant:   3.3 [CUED 28/04/05]";
char *hquant_vc_id = "$Id: HQuant.c,v 1.1.1.1 2005/05/12 10:52:54 jal58 Exp $";

/* 
   This program calculates a vector quantisation table from a
   sequence of training files.
*/ 

#include "HShell.h"
#include "HMem.h"
#include "HMath.h"
#include "HSigP.h"
#include "HAudio.h"
#include "HWave.h"
#include "HVQ.h"
#include "HParm.h"
#include "HLabel.h"
#include "HModel.h"
#include "HTrain.h"

/* ------------------- Trace Flags & Vars ------------------------ */

#define T_TOP     0001              /* basic progress reporting */
#define T_MEAN    0002              /* dump global mean and cov */
#define T_LOAD    0004              /* trace data loading */
#define T_SEGS    0010              /* list label segments */
#define T_CLUST   0020              /* dump clusters */
#define T_TAB     0040              /* dump vq table */

static int trace = 0;               /* trace level */
static ConfParam *cParm[MAXGLOBS];   /* configuration parameters */
static int nParm = 0;               /* total num params */

/* ---------------------- Global Variables ----------------------- */

#define DEF_NCLUST 256              /* default codebook size */

static BufferInfo info;             /* global observation format, etc. */
static int cbSizes[SMAX];           /* codebook sizes, per stream */
static short swidth[SMAX];          /* stream widths */
static TreeType tType = linTree;    /* codebook structure */
static char *vqfn = NULL;           /* filename for output VQ table */
static char *segLab = NULL;         /* segment label, if any */
static LabId segId  = NULL;         /* and its id */
static char *labDir = NULL;         /* label file directory */
static char *labExt = "lab";        /* label file extension */
static FileFormat dff=UNDEFF;       /* data file format */
static FileFormat lff=UNDEFF;       /* label file format */

static MemHeap iStack;             /* input buffer  */
static MemHeap dStack;             /* sequence stack */
static MemHeap cStack;             /* cluster stack */
Sequence dSeq[SMAX];               /* main data pools; one per stream */
ClusterSet *cs[SMAX];              /* stores vector clusters */
CovKind ck = NULLC;                /* determines distance metric */
Vector mean[SMAX];                 /* Global stream mean  */
Covariance cov[SMAX];              /* Global stream covariance  */
Boolean widthSet = FALSE;          /* true if width of any stream is set */

static Observation obs;             /* storage for observations  */
static Boolean globClustVar = FALSE;/*Output global variance of data to
                                      codebook in place of individual vars*/

/* ------------- Process Command Line and Check Data ------------ */

/* SetConfParms: set conf parms relevant to HQuant  */
void SetConfParms(void)
{
   int i;
   
   nParm = GetConfig("HQUANT", TRUE, cParm, MAXGLOBS);
   if (nParm>0) {
      if (GetConfInt(cParm,nParm,"TRACE",&i)) trace = i;
   }
}

/* InitStreamVars: set initial values of stream-indexed variables */
void InitStreamVars(void)
{
   int s;

   swidth[0] = cbSizes[0] = 0;
   for(s=1;s<SMAX;s++) {
      swidth[s] = 0;
      cbSizes[s] = DEF_NCLUST;
      mean[s] = NULL;
   }
}

void ReportUsage(void)
{
   printf("\nUSAGE: HQuant [options] vqFile trainFiles...\n\n" );
   printf(" Option                                       Default\n\n");
   printf(" -d      Use diagonal cov Mahalanobis         Euclidean\n");
   printf(" -f      Use full covariance Mahalanobis      Euclidean\n");
   printf(" -g      Output global covar to codebook      off\n");
   printf(" -l s    Set segment label to s               none\n");
   printf(" -n S N  Set codebook size for stream S to N  N=%d\n",DEF_NCLUST);
   printf(" -s N    Set number of streams to N           1\n");
   printf(" -t      Create tree-stuctured codebooks      linear\n");
   printf(" -w S N  Set width of stream S to N           default\n");
   PrintStdOpts("FGILX");
   printf("\n\n");
}

int main(int argc, char *argv[])
{
   char *datafn, *s;
   int stream = 0;
   void Initialise(char *datafn);
   void LoadFile(char *fn);
   void CalcMeanCov(Sequence seq[], int s);
   void ClusterVecs(Sequence seq[], int s);
   void WriteVQTable(ClusterSet *cs[], char *fn);  

   if(InitShell(argc,argv,hquant_version,hquant_vc_id)<SUCCESS)
      HError(2500,"HQuant: InitShell failed");

   InitMem();   InitLabel();
   InitMath();  InitSigP();
   InitWave();  InitAudio();
   InitVQ();    InitModel();
   if(InitParm()<SUCCESS)  
      HError(2500,"HQuant: InitParm failed");
   InitTrain();

   if (!InfoPrinted() && NumArgs() == 0)
      ReportUsage();
   if (NumArgs() == 0) Exit(0);
   SetConfParms();
   InitStreamVars();

   while (NextArg() == SWITCHARG) {
      s = GetSwtArg();
      if (strlen(s)!=1) 
         HError(2519,"HQuant: Bad switch %s; must be single letter",s);
      switch(s[0]){
      case 'd':
         if ( ck != NULLC) HError(2519,"HQuant: Specify one of -d or -f, not both");
         ck = INVDIAGC;
         break;
      case 'f':
         if ( ck != NULLC) HError(2519,"HQuant: Specify one of -d or -f, not both");
         ck = FULLC;
         break;
      case 'g':
         globClustVar = TRUE;
         break;
      case 'l':
         if (NextArg() != STRINGARG)
            HError(2519,"HQuant: Segment label expected");
         segLab = GetStrArg();
         break;
      case 'n':
         if (NextArg() != INTARG)
            HError(2519,"HQuant: Stream number expected");
         stream = GetChkedInt(1,SMAX,s);
         if (NextArg() != INTARG)
            HError(2519,"HQuant: Codebook size expected");
         cbSizes[stream]= GetChkedInt(1,32768,s);
         break;
      case 's':
         if (NextArg() != INTARG)
            HError(2519,"HQuant: Number of streams expected");
         swidth[0] = GetChkedInt(1,SMAX,s);
         break;
      case 't':
         tType = binTree;
         break;
      case 'w':
         if (NextArg() != INTARG)
            HError(2519,"HQuant: Stream number expected");
         stream = GetChkedInt(1,SMAX,s);
         if(swidth[0] < stream) swidth[0] = stream;
         widthSet = TRUE;
         if (NextArg() != INTARG)
            HError(2519,"HQuant: Stream width expected");
         swidth[stream]= GetChkedInt(1,256,s);
         break;
      case 'F':
         if (NextArg() != STRINGARG)
            HError(2519,"HQuant: Data File format expected");
         if((dff = Str2Format(GetStrArg())) == ALIEN)
            HError(-2589,"HQuant: Warning ALIEN Data file format set");
         break;
      case 'G':
         if (NextArg() != STRINGARG)
            HError(2519,"HQuant: Label File format expected");
         if((lff = Str2Format(GetStrArg())) == ALIEN)
            HError(-2589,"HQuant: Warning ALIEN Label file format set");
         break;
      case 'I':
         if (NextArg() != STRINGARG)
            HError(2519,"HQuant: MLF file name expected");
         LoadMasterFile(GetStrArg());
         break;
      case 'L':
         if (NextArg()!=STRINGARG)
            HError(2519,"HQuant: Label file directory expected");
         labDir = GetStrArg();
         break;
      case 'T':
         if (NextArg() != INTARG)
            HError(2519,"HQuant: Trace value expected");
         trace = GetChkedInt(0,077,s); 
         break;
      case 'X':
         if (NextArg()!=STRINGARG)
            HError(2519,"HQuant: Label file extension expected");
         labExt = GetStrArg();
         break;
      default:
         HError(2519,"HQuant: Unknown switch %s",s);
      }
   }
   if (NextArg()!=STRINGARG)
      HError(2519,"HQuant: Output VQ table file name expected");
   vqfn = GetStrArg();

   if (NextArg()!=STRINGARG)
      HError(2519,"HQuant: Training data file name expected");
   datafn = GetStrArg();
   Initialise(datafn);
   LoadFile(datafn);
   while (NumArgs()>0) {
      if (NextArg()!=STRINGARG) 
         HError(2519,"HQuant: Training data file name expected");
      datafn = GetStrArg();
      LoadFile(datafn);
   }
   
   for (stream=1;stream<=swidth[0];stream++){
      if (trace&T_TOP)
         printf("%s-clustering data for stream %d (width %d)\n",
                (tType==linTree)?"Flat":"Tree",stream,swidth[stream]);
      CalcMeanCov(dSeq,stream);
      ClusterVecs(dSeq,stream);
   }
   WriteVQTable(cs,vqfn);
   Exit(0);
   return (0);          /* never reached -- make compiler happy */
}


/* Determine global mean and covariance of data in sequence s */
void CalcMeanCov(Sequence seq[], int s)
{