hbuild.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: HBuild.c:  Word-Lattice Building                  */
/* ----------------------------------------------------------- */

char *hbuild_version = "!HVER!HBuild:   3.3 [CUED 28/04/05]";
char *hbuild_vc_id = "$Id: HBuild.c,v 1.1.1.1 2005/05/12 10:52:53 jal58 Exp $";

/* The HBuild program takes input files in a number of different
   formats and constructs suitable HTK word lattice files.

   The formats currently supported by HBuild include:

   a) Bigrams in either ARPA/Lincol-Labs format or HTK matrix format
   b) HTK Multi-Level lattices
   c) Word Lists for simple loops
   d) ARPA word-pair grammars (Resource Management style)
*/

/* Trace Flags */
#define T_TOP        0001    /* Top Level tracing */

#include "HShell.h" /* HMM ToolKit Modules */
#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 "HUtil.h" 
#include "HDict.h"
#include "HNet.h"
#include "HLM.h"

typedef enum {unknown, wordLoop, boBiGram, matBiGram, multiLat, wordPair} BuildType;

static int trace     = 0;           /* Trace flags */

static LabId enterId;               /* id of !ENTRY label in ngram */
static LabId exitId;                /* id of !EXIT label in ngram */
static LabId bStartId=NULL;         /* id of start bracket */
static LabId bEndId=NULL;           /* id of end bracket */

static LabId unknownId;             /* id of unknown label in ngram */
static Boolean zapUnknown = FALSE;  /* zap unknown symbols from bigram */

MemHeap buildStack;

/* ---------------- Configuration Parameters --------------------- */

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

/* ---------------- Process Command Line ------------------------- */

/* SetConfParms: set conf parms relevant to this tool */
void SetConfParms(void)
{
   int i;

   nParm = GetConfig("HBUILD", TRUE, cParm, MAXGLOBS);
   if (nParm>0){
      if (GetConfInt(cParm,nParm,"TRACE",&i)) trace = i;
   }
}

void ReportUsage(void)
{
   printf("\nUSAGE: HBuild [options] wordList latFile\n\n");
   printf(" Option                                       Default\n\n");
   printf(" -b      binary lattice output                ASCII\n");
   printf(" -m s    load matrix bigram from s            off\n");
   printf(" -n s    load back-off bigram from s          off\n");
   printf(" -s s1 s2 s1/s2 are bigram start/end labels   !ENTER !EXIT\n");
   printf(" -t s1 s2 bracket word-loop/pair with s1 s2   off\n");
   printf(" -u s    set unknown symbol to s              !NULL\n");
   printf(" -w s    load word-pair grammar from s        off\n");
   printf(" -x s    load multi-level lattice from s      off\n");
   printf(" -z      ignore ngrams with unknown symbol    off\n");
   PrintStdOpts(""); 
   printf("\n\n");
}

int main(int argc, char *argv[])
{
   char *wordListFn,*latFn,*ipFn=NULL;
   LModel *bigramLm;
   BuildType bType = unknown;
   Boolean saveLatBin = FALSE;
   LatFormat format = HLAT_LMLIKE;
   Lattice *lat,*ipLat;
   Vocab voc;
   char  *s;

   Lattice *ProcessWordLoop(MemHeap *latHeap, Vocab *voc);
   Lattice *ProcessBiGram(MemHeap *latHeap, Vocab *voc, LModel *biLM);
   void SaveLattice(Lattice *lat, char *latFn, LatFormat format);
   Lattice *LoadLattice(MemHeap *latHeap, char *latFn, Vocab *voc,
                        Boolean shortArc);
   Lattice *ProcessWordPair(MemHeap *latHeap, Vocab *voc, char *fn);

   if(InitShell(argc,argv,hbuild_version,hbuild_vc_id)<SUCCESS)
      HError(3000,"HBuild: InitShell failed");
   InitMem();   InitLabel();
   InitMath();  
   InitDict();  InitNet();  
   InitLM();

   CreateHeap(&buildStack, "HBuild Stack",  MSTAK, 1, 0.0, 100000, LONG_MAX );

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

   enterId=GetLabId("!ENTER",TRUE);   /* All sentences should or are coerced */
   exitId=GetLabId("!EXIT",TRUE);     /*  to start enterId and end exitId */
   unknownId=GetLabId("!NULL",TRUE);  /* Name for words not in list */

   while (NextArg() == SWITCHARG) {
      s = GetSwtArg();
      if (strlen(s)!=1) 
         HError(3019,"HBuild: Bad switch %s; must be single letter",s);
      switch(s[0]){
      case 'b':
         saveLatBin = TRUE; break;    
      case 'm':
         if (bType != unknown)
            HError(3019,"HBuild: Can only specifiy one of -m, -n, -w, -x");
         bType = matBiGram;
         if (NextArg()!=STRINGARG)
            HError(3019,"HBuild: Matrix Bigram file name expected");
         ipFn = GetStrArg(); 
         break;
      case 'n':
         if (bType != unknown)
            HError(3019,"HBuild: Can only specifiy one of -m, -n, -w, -x");
         bType = boBiGram;
         if (NextArg()!=STRINGARG)
            HError(3019,"HBuild: Back-off Bigram file name expected");
         ipFn = GetStrArg(); 
         break;
      case 's':
         if (NextArg() != STRINGARG)
            HError(3019,"HBuild: Bigram ENTER label name expected");
         enterId=GetLabId(GetStrArg(),TRUE);
         if (NextArg() != STRINGARG)
            HError(3019,"HBuild: Bigram EXIT label name expected");
         exitId=GetLabId(GetStrArg(),TRUE);
         break;
      case 't':
         if (NextArg() != STRINGARG)
            HError(3019,"HBuild: Bracket start label name expected");
         bStartId=GetLabId(GetStrArg(),TRUE);
         if (NextArg() != STRINGARG)
            HError(3019,"HBuild: Bracket end label name expected");
         bEndId=GetLabId(GetStrArg(),TRUE);
         break;
      case 'u':
         if (NextArg() != STRINGARG)
            HError(3019,"HBuild: Unknown label name expected");
         unknownId=GetLabId(GetStrArg(),TRUE);
         break;
      case 'w':
         if (bType != unknown)
            HError(3019,"HBuild: Can only specifiy one of -m, -n, -w, -x");
         bType = wordPair;
         if (NextArg()!=STRINGARG)
            HError(3019,"HBuild: Word pair grammar file name expected");
         ipFn = GetStrArg(); 
         break;
      case 'x':
         if (bType != unknown)
            HError(3019,"HBuild: Can only specifiy one of -m, -n, -w, -x");
         bType = multiLat;
         if (NextArg()!=STRINGARG)
            HError(3019,"HBuild: Multi-level lattice file name expected");
         ipFn = GetStrArg(); 
         break;
      case 'z':
         zapUnknown = TRUE; break;    
      case 'T':
         trace = GetChkedInt(0,511,s); break;
      default:
         HError(3019,"HBuild: Unknown switch %s",s);
      }
   } 
   if (NextArg()!=STRINGARG)
      HError(3019,"HBuild: Word List file name expected");
   wordListFn = GetStrArg();
   if (NextArg()!=STRINGARG)
      HError(3019,"HBuild: output lattice file name expected");
   latFn = GetStrArg();
   if (bType == unknown) bType = wordLoop;
   if (saveLatBin) format |= HLAT_LBIN;   
   /* Read the word-list into a Vocab data structure */
   InitVocab(&voc);
   if(ReadDict(wordListFn, &voc)<SUCCESS)
      HError(3013,"HBuild: ReadDict failed");
   switch (bType) {
   case matBiGram:
      if (trace & T_TOP)
         printf("Reading bigram from file %s\n",ipFn);
      bigramLm = ReadLModel(&gstack, ipFn);
      if (bigramLm->type != matBigram)
         HError(3030,"HBuild: File specified is not a matrix bigram");
      lat = ProcessBiGram(&gstack,&voc,bigramLm);
      SaveLattice(lat,latFn,format);
      break;
   case boBiGram:
      if (trace & T_TOP)
         printf("Reading bigram from file %s\n",ipFn);
      bigramLm = ReadLModel(&gstack, ipFn);
      if (bigramLm->type != boNGram)
         HError(3030,"HBuild: File specified is not a back-off bigram");
      lat = ProcessBiGram(&gstack,&voc,bigramLm);
      SaveLattice(lat,latFn,format);
      break;
   case multiLat:
      if (trace & T_TOP)
         printf("Reading input lattice from file %s\n",ipFn);
      ipLat = LoadLattice(&buildStack,ipFn,&voc,FALSE);
      if (ipLat->subList!=NULL) {
         if (trace & T_TOP)
            printf("Expanding multi-level lattice\n");
         lat = ExpandMultiLevelLattice(&buildStack,ipLat,&voc);
      }
      else
         lat = ipLat;
      SaveLattice(lat,latFn,format);
      break;
   case wordLoop:
      if (trace & T_TOP)
         printf("Building word loop\n");
      lat = ProcessWordLoop(&gstack,&voc);
      SaveLattice(lat,latFn,format);
      break;
   case wordPair:
      lat = ProcessWordPair(&gstack,&voc,ipFn);
      SaveLattice(lat,latFn,format);
      break;
   default:
      HError(3001,"Only Bigram LMs / multiLats currently implemented");
   }
   Exit(0);
   return (0);          /* never reached -- make compiler happy */
}

/* Save a lattice to a file latFn  */
void SaveLattice(Lattice *lat, char *latFn, LatFormat format)
{
   FILE *latf;
   Boolean isPipe;

   if (trace & T_TOP)
      printf("Saving lattice to file %s\n",latFn);
   if ( (latf = FOpen(latFn,NetOFilter,&isPipe)) == NULL)
      HError(3011,"SaveLattice : Cannot create new lattice file  %s",latFn);
   if(WriteLattice(lat,latf,format)<SUCCESS)
      HError(3011,"SaveLattice : Cannot create new lattice file  %s",latFn);
   FClose(latf,isPipe);
}

/* Load a lattice from file latFn  */
Lattice *LoadLattice(MemHeap *latHeap, char *latFn, Vocab *voc,
                     Boolean shortArc)
{
   FILE *latf;
   Boolean isPipe;
   Lattice *lat;

   if ( (latf = FOpen(latFn,NetFilter,&isPipe)) == NULL)
      HError(3010,"LoadLattice : Cannot open lattice file %s",latFn);
   if((lat = ReadLattice(latf,latHeap,voc,shortArc,FALSE))==NULL)
      HError(3010,"LoadLattice : ReadLattice failed");
   FClose(latf,isPipe);
   return lat;
}

Lattice *ProcessWordLoop(MemHeap *latHeap, Vocab *voc)
{
   int nNode,nArc;
   LNode *ln;
   LArc *la;
   Word wd;
   Lattice *lat;
   int i;

   nNode = voc->nwords+4; 
   nArc =  voc->nwords*2 + 3;
   
   lat = NewLattice(latHeap,nNode,nArc);
   lat->voc = voc;
   lat->lmscale = 1.0; lat->wdpenalty = 0.0; 

   /* fill in start/end/loop word entries with !NULL */
   wd = voc->nullWord;
   ln = lat->lnodes; ln->word = wd; ln->n=0; ln->v=0;
   ln = lat->lnodes+1; ln->word = wd; ln->n=0; ln->v=0;
   ln = lat->lnodes+nNode-1; ln->word = wd; ln->n=0; ln->v=0;
   ln = lat->lnodes+nNode-2; ln->word = wd; ln->n=0; ln->v=0;

   ln = lat->lnodes+2;
   for (i = 0; i< VHASHSIZE; i++)
      for ( wd = voc->wtab[i]; wd != NULL; wd = wd->next ) 
         if ((wd != voc->nullWord) && (wd != voc->subLatWord)) {
            ln->word = wd;
            ln++;
         }

   la =lat->larcs;
   la->start = lat->lnodes;  
   la->end = lat->lnodes+1; 
   la->lmlike = 0.0;
   la = lat->larcs+1;
   la->start = lat->lnodes+nNode-2;  
   la->end = lat->lnodes+nNode-1;
   la->lmlike = 0.0;
   la = lat->larcs+2;
   la->start = lat->lnodes+nNode-2;   
   la->end = lat->lnodes+1;
   la->lmlike = 0.0;

   la = lat->larcs+3;
   for (i = 0; i < voc->nwords; i++) {
      la->start = lat->lnodes+1;
      la->end = lat->lnodes+2+i;
      la->lmlike = log(1.0/(float) (voc->nwords));
      la++;
   }
   for (i = 0; i < voc->nwords; i++) {
      la->start = lat->lnodes+2+i;
      la->end = lat->lnodes+nNode-2;
      la->lmlike = 0.0;
      la++;
   }
   /* finally overwrite start/end !NULL words if sil at start/end */
   if (bStartId != NULL) {
      wd = GetWord(voc,bStartId,TRUE);
      ln = lat->lnodes; ln->word = wd;
      wd = GetWord(voc,bEndId,TRUE);
      ln = lat->lnodes+nNode-1; ln->word = wd;
   }
   return lat;
}


/*ProcessBoBiGram: Convert back-off bigram in nLM into lattice */ 
Lattice *ProcessBoBiGram(MemHeap *latHeap, Vocab *voc, NGramLM *nLM)
{
   int nNode,nArc;
   NEntry *ne;
   SEntry *se;
   Word wd,fromWd,toWd;
   LNode *ln,*fromNode,*toNode;
   LArc *la;

   lmId ndx[NSIZE+1];  
   int i,j,k;