/* ReplaceUnigrams: replace unigrams in lm with ones from fn */
void ReplaceUnigrams(char *fn, BackOffLM *lm)
{
float prob;
SMEntry *se;
NameId wdid;
int ndx,nItem;
FLEntry *root;
char buf[MAXSTRLEN];
Source src;
nItem = 0;
root = &lm->root;
if(InitSource(fn,&src,NoFilter)<SUCCESS)
HError(16910,"ReplaceUnigrams: Can't open file %s", fn);
do {
if (!ReadFloat(&src,&prob,1,FALSE)) break;
if (!ReadRawString(&src,buf)) break; /* or ReadString if HTK escaped */
if ((wdid = GetNameId(lm->htab,buf,FALSE))==NULL) {
printf("skipping '%s'\n", buf);
continue;
}
ndx = LM_INDEX(wdid); se = root->sea+ndx-1;
if (se->ndx!=ndx && (se=FindSE(root->sea,0,root->nse,ndx))==NULL) {
printf("ignoring '%s'\n", buf);
continue;
}
se->prob = exp(prob*LN10);
nItem++;
} while(SkipLine(&src));
CloseSource(&src);
if (trace&T_TOP) {
printf("Replaced %d unigrams from %s\n",nItem,uniFn); fflush(stdout);
}
}
int CTwain::DropToState(int nS, HWND hwnd)
{
hwnd = DefWnd(hwnd);
while (nState > nS) {
switch (nState) {
case TRANSFERRING:
if (!EndXfer()) return FALSE;
break;
case TRANSFER_READY:
if (!CancelXfers()) return FALSE;
break;
case SOURCE_ENABLED:
if (!DisableSource()) return FALSE;
break;
case SOURCE_OPEN:
if (!CloseSource()) return FALSE;
break;
case SOURCE_MANAGER_OPEN:
if (!CloseSourceManager(hwnd)) return FALSE;
break;
case SOURCE_MANAGER_LOADED:
if (!UnloadSourceManager()) return FALSE;
break;
default:
ATLASSERT(FALSE);
} // switch
} // while
return TRUE;
}
/* EXPORT ReadLModel: Determine LM type and then read-in */
LModel *ReadLModel(MemHeap *heap,char *fn)
{
LModel *lm;
LMType type;
char buf[MAXSTRLEN+1];
int i;
lm=(LModel*)New(heap,sizeof(LModel));
lm->heap=heap;
lm->name=CopyString(heap,fn);
if(InitSource(fn,&source,LangModFilter)<SUCCESS)
HError(8110,"ReadLModel: Can't open file %s", fn);
type=boNGram;i=0;
do {
if (i++==1000) {
type=matBigram;
break;
}
GetInLine(buf);
}
while (strcmp(buf, "\\data\\")!=0);
CloseSource(&source);
lm->type=type;
switch(type) {
case boNGram:
ReadBoNGram(lm,fn);
break;
case matBigram:
ReadMatBigram(lm,fn);
break;
}
return(lm);
}
/* ReadBoNGram: read and store WSJ/DP format ngram */
static void ReadBoNGram(LModel *lm,char *fn)
{
NGramLM *nglm;
int i,j,k,counts[NSIZE+1];
Boolean ngBin[NSIZE+1];
char buf[MAXSTRLEN+1],syc[64];
char ngFmtCh;
if (trace&T_TIO)
printf("\nBOffB "),fflush(stdout);
if(InitSource(fn,&source,LangModFilter)<SUCCESS)
HError(8110,"ReadBoNGram: Can't open file %s", fn);
GetInLine(buf);
SyncStr(buf,"\\data\\");
for (i=1;i<=NSIZE;i++) counts[i]=0;
for (i=1;i<=NSIZE;i++) {
GetInLine(buf);
if (sscanf(buf, "ngram %d%c%d", &j, &ngFmtCh, &k)!=3 && i>1)
break;
if (i!=j || k==0)
HError(8150,"ReadBoNGram: %dGram count missing (%s)",i,buf);
switch (ngFmtCh) {
case '=':
ngBin[j] = FALSE;
break;
case '~':
ngBin[j] = TRUE;
break;
default:
HError (9999, "ReadARPALM: unknown ngram format type '%c'", ngFmtCh);
}
counts[j]=k;
}
if (ngBin[1])
HError (8113, "ReadARPALM: unigram must be stored as text");
nglm=CreateBoNGram(lm,counts[1],counts);
for (i=1;i<=nglm->nsize;i++) {
sprintf(syc,"\\%d-grams:",i);
SyncStr(buf,syc);
ReadNGrams(nglm,i,nglm->counts[i], ngBin[i]);
}
SyncStr(buf,"\\end\\");
CloseSource(&source);
if (trace&T_TIO) {
printf("\n NEntry==%d ",nglm->counts[0]);
for(i=1;i<=nglm->nsize;i++)
printf(" %d-Grams==%d",i,nglm->counts[i]);
printf("\n\n");
fflush(stdout);
}
}
int CTwain::CloseSourceManager(HWND hwnd)
{
CloseSource(); // close source if open
if (nState >= SOURCE_MANAGER_OPEN) {
TW_INT32 hwnd32 = (TW_INT32)(int)(DefWnd(hwnd));
SM(DG_CONTROL, DAT_PARENT, MSG_CLOSEDSM, &hwnd32);
if (nState >= SOURCE_MANAGER_OPEN && bTrace) ATLTRACE(_T("TWAIN:CLOSEDSM failed.\n"));
}
return (nState < SOURCE_MANAGER_OPEN);
}
static int trace = 0; #define T_TOP 0001 /* top level tracing */ #define T_HDR 0002 /* show header processing */ #define T_HASH 0004 /* hash table tracing */
/* EXPORT->ReadDict: read and store a dictionary definition */
ReturnStatus ReadDict(char *dictFn, Vocab *voc)
{
LabId labels[MAXPHONES+4];
Source src;
Word word;
float prob;
int nphones;
ReturnStatus ret;
if(InitSource(dictFn,&src,DictFilter)<SUCCESS){
HRError(8010,"ReadDict: Can't open file %s", dictFn);
return(FAIL);
}
if (trace&T_TOP)
printf("\nLoading Dictionary from %s\n",dictFn);
if((ret=ReadDictWord(&src,labels,&prob, &nphones))<SUCCESS){
CloseSource(&src);
HRError(8013,"ReadDict: Dict format error in first entry");
return(FAIL);
}
while(nphones>=0){
word = GetWord(voc,labels[0],TRUE);
if (labels[1]==NULL) labels[1]=labels[0];
if (labels[1]->name[0]==0) labels[1]=NULL;
if (voc->nullWord->wordName == word->wordName)
HRError(-8013,"ReadDict: !NULL entry contains pronunciation");
NewPron(voc,word,nphones,labels+2,labels[1],prob);
if((ret=ReadDictWord(&src,labels,&prob, &nphones))<SUCCESS){
HRError(8013,"ReadDict: Dict format error");
return(FAIL);
}
}
CloseSource(&src);
if (trace&T_DIC)
ShowDict(voc);
if (trace&T_TOP)
printf("Dictionary loaded from %s with %d words and %d prons\n\n",
dictFn,voc->nwords,voc->nprons);
return(SUCCESS);
}
/* EXPORT->LoadStatsFile: load the statistics file output by HERest */
void LoadStatsFile(char *statfile,HMMSet *hset,Boolean otrace)
{
Source src;
char hname[256];
int i,idx,count,N,lnum = 0;
float x;
HMMDef *hmm;
MLink ml;
LabId hmmId;
double occSum = 0.0;
long occN = 0;
StateInfo *si;
Boolean bin=FALSE;
if(InitSource(statfile,&src,NoFilter)<SUCCESS)
HError(7210,"LoadStatsFile: Can't open file %s", statfile);
while(ReadInt(&src,&idx,1,bin)) {
++lnum;
if (!ReadString(&src,hname) || !ReadInt(&src,&count,1,bin))
HError(7250,"LoadStatsFile: Format error in file %s line %d",
statfile,lnum);
/* look up hname and find num states N */
if ((hmmId = GetLabId(hname,FALSE))==NULL)
HError(7251,"LoadStatsFile: unknown name %s at line %d",
hname,lnum);
if ((ml = FindMacroName(hset,'l',hmmId))==NULL)
HError(7251,"LoadStatsFile: unknown model %s at line %d",
hname,lnum);
hmm = (HMMDef *) ml->structure;
N = hmm->numStates;
for (i=2; i<N; i++) {
if (!ReadFloat(&src,&x,1,bin))
HError(7250,"LoadStatsFile: Float format error file %s line %d\n",
statfile,lnum);
si = hmm->svec[i].info;
si->stateCounter = count;/* load the # of times the state occurred */
memcpy(&(si->hook),&x,sizeof(float)); /* !! */
occSum += x; ++occN;
}
}
CloseSource(&src);
if (otrace || (trace & T_OCC)) {
printf(" Stats loaded for %d models\n",lnum);
printf(" Mean Occupation Count = %f\n",occSum/occN);
fflush(stdout);
}
}
// destruktor
CDataSourcesManager::~CDataSourcesManager()
{
// kody: zavreni vsech datovych zdroju (kvuli Ferdovi)
// ***
saveSourcesTab();
for (int i = 0; i < SourcesTab.GetSize(); i++)
{
try
{
if(isSourceConnected(i))
CloseSource(i);
}
catch(...) {}
delete SourcesTab[i];
}
// ***
}
void FCDProThread::run()
{
if ( !OpenSource(fcd_traits<Pro>::alsaDeviceName) )
{
qCritical() << "FCDThread::run: cannot open FCD sound card";
return;
}
// TODO: fallback to original fcd
m_running = true;
while(m_running)
{
if (work(fcd_traits<Pro>::convBufSize) < 0)
{
break;
}
}
CloseSource();
}
void CTwain::CloseDsRequest(void)
{
CloseSource();
}
/* SetConfParms: set conf parms relevant to HCompV */
void SetConfParms(void)
{
int i;
Boolean b;
double f;
char buf[MAXSTRLEN];
nParm = GetConfig("HEREST", TRUE, cParm, MAXGLOBS);
if (nParm>0) {
if (GetConfInt(cParm,nParm,"TRACE",&i)) trace = i;
if (GetConfFlt(cParm,nParm,"VARFLOORPERCENTILE",&f)) varFloorPercent = f;
if (GetConfBool(cParm,nParm,"SAVEBINARY",&b)) saveBinary = b;
if (GetConfBool(cParm,nParm,"BINARYACCFORMAT",&b)) ldBinary = b;
/* 2-model reestimation alignment model set */
if (GetConfStr(cParm,nParm,"ALIGNMODELMMF",buf)) {
strcpy(al_hmmMMF,buf); al_hmmUsed = TRUE;
}
if (GetConfStr(cParm,nParm,"ALIGNHMMLIST",buf)) {
strcpy(al_hmmLst,buf); al_hmmUsed = TRUE;
}
/* allow multiple individual model files */
if (GetConfStr(cParm,nParm,"ALIGNMODELDIR",buf)) {
strcpy(al_hmmDir,buf); al_hmmUsed = TRUE;
}
if (GetConfStr(cParm,nParm,"ALIGNMODELEXT",buf)) {
strcpy(al_hmmExt,buf); al_hmmUsed = TRUE;
}
if (GetConfStr(cParm,nParm,"ALIGNXFORMEXT",buf)) {
xfInfo.alXFormExt = CopyString(&hmmStack,buf);
}
if (GetConfStr(cParm,nParm,"ALIGNXFORMDIR",buf)) {
xfInfo.alXFormDir = CopyString(&hmmStack,buf);
}
if (GetConfStr(cParm,nParm,"INXFORMMASK",buf)) {
xfInfo.inSpkrPat = CopyString(&hmmStack,buf);
}
if (GetConfStr(cParm,nParm,"PAXFORMMASK",buf)) {
xfInfo.paSpkrPat = CopyString(&hmmStack,buf);
}
if (GetConfStr(cParm,nParm,"LABFILEMASK",buf)) {
labFileMask = (char*)malloc(strlen(buf)+1);
strcpy(labFileMask, buf);
}
if (GetConfStr(cParm,nParm,"UPDATEMODE",buf)) {
if (!strcmp (buf, "DUMP")) updateMode = UPMODE_DUMP;
else if (!strcmp (buf, "UPDATE")) updateMode = UPMODE_UPDATE;
else if (!strcmp (buf, "BOTH")) updateMode = UPMODE_BOTH;
else HError(2319, "Unknown UPDATEMODE specified (must be DUMP, UPDATE or BOTH)");
}
}
}
void ReportUsage(void)
{
printf("\nUSAGE: HERest [options] hmmList dataFiles...\n\n");
printf(" Option Default\n\n");
printf(" -a Use an input linear transform off\n");
printf(" -c f Mixture pruning threshold 10.0\n");
printf(" -d s dir to find hmm definitions current\n");
printf(" -h s set output speaker name pattern *.%%%%%%\n");
printf(" to s, optionally set input and parent patterns\n");
printf(" -l N set max files per speaker off\n");
printf(" -m N set min examples needed per model 3\n");
printf(" -o s extension for new hmm files as src\n");
printf(" -p N set parallel mode to N off\n");
printf(" -r Enable Single Pass Training... \n");
printf(" ...using two parameterisations off\n");
printf(" -s s print statistics to file s off\n");
printf(" -t f [i l] set pruning to f [inc limit] inf\n");
printf(" -u tmvwap update t)rans m)eans v)ars w)ghts tmvw\n");
printf(" a)daptation xform p)rior used \n");
printf(" s)semi-tied xform \n");
printf(" -v f set minimum variance to f 0.0\n");
printf(" -w f set mix weight floor to f*MINMIX 0.0\n");
printf(" -x s extension for hmm files none\n");
printf(" -z s Save all xforms to TMF file s TMF\n");
PrintStdOpts("BEFGHIJKLMSTX");
printf("\n\n");
}
void SetuFlags(void)
{
char *s;
s=GetStrArg();
uFlags=(UPDSet) 0;
while (*s != '\0')
switch (*s++) {
case 't': uFlags = (UPDSet) (uFlags+UPTRANS); break;
case 'm': uFlags = (UPDSet) (uFlags+UPMEANS); break;
case 'v': uFlags = (UPDSet) (uFlags+UPVARS); break;
case 'w': uFlags = (UPDSet) (uFlags+UPMIXES); break;
case 's': uFlags = (UPDSet) (uFlags+UPSEMIT); break;
case 'a': uFlags = (UPDSet) (uFlags+UPXFORM); break;
case 'p': uFlags = (UPDSet) (uFlags+UPMAP); break;
default: HError(2320,"SetuFlags: Unknown update flag %c",*s);
break;
}
}
/* ScriptWord: return next word from script */
char *ScriptWord(FILE *script, char *scriptBuf)
{
int ch,qch,i;
i=0; ch=' ';
while (isspace(ch)) ch = fgetc(script);
if (ch==EOF) {
scriptBuf=NULL;
return NULL;
}
if (ch=='\'' || ch=='"'){
qch = ch;
ch = fgetc(script);
while (ch != qch && ch != EOF) {
scriptBuf[i++] = ch;
ch = fgetc(script);
}
if (ch==EOF)
HError(5051,"ScriptWord: Closing quote missing in script file");
} else {
do {
scriptBuf[i++] = ch;
ch = fgetc(script);
}while (!isspace(ch) && ch != EOF);
}
scriptBuf[i] = '\0';
return scriptBuf;
}
void CheckUpdateSetUp()
{
AdaptXForm *xf;
VOID
__cdecl main (
IN int argc,
IN char **argv
)
{
//
// Initialize
//
InitParse ();
ParseArgs (argc, argv);
InitializeListHead (&VerifyRef);
if (!Source) {
HelpAndExit (NULL);
}
printf ("ACPI Souce Language Assembler Version 0.1\n");
printf ("Copyright (C) Microsoft Corp 1996. All rights reserved.\n");
fflush (stdout);
//
// Build top level scope & ASL data package for which image
// can be built in
//
DataImage = AllocAl();
DataImage->Name = AllocName();
DataImage->Flags |= F_AMLPACKAGE | F_PVARIABLE;
DataImage->Term = &ImageTerm;
DataImage->Parent = DataImage;
DataImage->DataType = TypeRoot;
InitializeListHead(&DataImage->FixedList);
InitializeListHead(&DataImage->u1.VariableList);
DataImage->Name->NameSeg = (ULONG) '\\';
AlLoc = DataImage;
//
// Parse source
//
while (Source) {
ParseSource ();
CloseSource ();
}
if (Verbose > 1) {
printf ("NameSpaceDump:\n");
DumpNameSpace (DataImage->Name, 0);
//not valid anymore...
//if (Verbose > 2) {
// DumpImage();
// printf ("AMLSize %d\n", AMLSize);
//}
}
//
// Enable AML for data image, and calculate package sizes
//
DataImage->Flags |= F_AMLENCODE;
WriteDefinitionBlocks (DataImage);
//
// Remove AML package size from image Al
//
DataImage->Flags &= ~F_AMLENCODE;
DataImage->u.Data.Length = 0;
//
// ...
//
Terminate ();
}
int main(int argc, char *argv[])
{
Source src;
int tmpInt;
float tmpFlt;
char *accfn, *s;
void Initialise(char *hmmListFn);
void Interpolate(void);
void UpdateModels(void);
void MakeWtAccLists(void);
void AttachWtAccLists(void);
void StatReport(void);
if(InitShell(argc,argv,hsmooth_version,hsmooth_vc_id)<SUCCESS)
HError(2400,"HSmooth: InitShell failed");
InitMem(); InitLabel();
InitMath(); InitSigP();
InitWave(); InitAudio();
InitVQ(); InitModel();
if(InitParm()<SUCCESS)
HError(2400,"HSmooth: InitParm failed");
InitTrain(); InitUtil();
if (!InfoPrinted() && NumArgs() == 0)
ReportUsage();
if (NumArgs() == 0) Exit(0);
SetConfParms();
CreateHeap(&hmmStack,"HmmStore", MSTAK, 1, 1.0, 50000, 500000);
CreateHMMSet(&hset,&hmmStack,TRUE);
while (NextArg() == SWITCHARG) {
s = GetSwtArg();
if (strlen(s)!=1)
HError(2419,"HSmooth: Bad switch %s; must be single letter",s);
switch(s[0]){
case 'b':
epsilon = GetChkedFlt(0.0,1.0,s); break;
case 'c':
maxStep = GetChkedInt(1,1000,s); break;
case 'd':
if (NextArg()!=STRINGARG)
HError(2419,"HSmooth: HMM definition directory expected");
hmmDir = GetStrArg(); break;
case 'e':
if (NextArg()!=STRINGARG)
HError(2419,"HSmooth: HMM definition directory expected");
newDir = GetStrArg(); break;
case 'm':
minEgs = GetChkedInt(1,1000,s); break;
case 'o':
if (NextArg()!=STRINGARG)
HError(2419,"HSmooth: HMM file extension expected");
newExt = GetStrArg(); break;
case 's':
stats = TRUE;
if (NextArg()!=STRINGARG)
HError(2419,"HSmooth: Stats file name expected");
statFN = GetStrArg(); break;
case 'u':
SetuFlags(); break;
case 'v':
minVar = GetChkedFlt(0.0,10.0,s); break;
case 'w':
mixWeightFloor = MINMIX * GetChkedFlt(0.0,10000.0,s);
break;
case 'x':
if (NextArg()!=STRINGARG)
HError(2419,"HSmooth: HMM file extension expected");
hmmExt = GetStrArg(); break;
case 'B':
saveBinary=TRUE;
break;
case 'H':
if (NextArg() != STRINGARG)
HError(2419,"HSmooth: HMM macro file name expected");
AddMMF(&hset,GetStrArg());
break;
case 'M':
if (NextArg()!=STRINGARG)
HError(2419,"HSmooth: Output macro file directory expected");
newDir = GetStrArg();
break;
case 'T':
trace = GetChkedInt(0,0100000,s); break;
default:
HError(2419,"HSmooth: Unknown switch %s",s);
}
}
if (NextArg() != STRINGARG)
HError(2419,"HSmooth: file name of HMM list expected");
Initialise(GetStrArg());
do {
if (NextArg()!=STRINGARG)
HError(2419,"HSmooth: accumulator file name expected");
accfn = GetStrArg();
src=LoadAccs(&hset,accfn,uFlags);
ReadFloat(&src,&tmpFlt,1,ldBinary);
totalPr += (LogDouble)tmpFlt;
ReadInt(&src,&tmpInt,1,ldBinary);
totalT += tmpInt;
CloseSource(&src);
nBlk++;
MakeWtAccLists();
} while (NumArgs()>0);
AttachWtAccLists();
Interpolate();
if (stats) StatReport();
UpdateModels();
Exit(0);
return (0); /* never reached -- make compiler happy */
}
/* ReadBigram: load a bigram from given file */
static void ReadMatBigram(LModel *lm,char *fn)
{
Vector vec;
char buf[132];
int P,p,j;
float sum,x;
LabId id;
MatBiLM *matbi;
if (trace&T_TIO)
printf("\nMB "),fflush(stdout);
if(InitSource(fn,&source,LangModFilter)<SUCCESS)
HError(8110,"ReadMatBigram: Can't open file %s", fn);
vec = CreateVector(&gcheap,MAX_LMID);
ReadLMWord(buf);SkipWhiteSpace(&source);
id=GetLabId(buf,TRUE);
P = ReadRow(vec);
if (P<=0 || P >MAX_LMID)
HError(8151,"ReadMatBigram: First row invalid (%d entries)",P);
matbi=CreateMatBigram(lm,P);
matbi->wdlist[1] = id;
for (p=1;p<=P;p++) matbi->bigMat[1][p]=vec[p];
id->aux=(Ptr) 1;
Dispose(&gcheap,vec);
for (sum=0.0, j=1; j<=P; j++) {
x = matbi->bigMat[1][j];
if (x<0)
HError(8151,"ReadMatBigram: In bigram, entry %d for %s is -ve (%e)",
j,buf,x);
sum += x;
matbi->bigMat[1][j]=((x<MINLARG)?LZERO:log(x));
}
if (sum < 0.99 || sum > 1.01)
HError(-8151,"ReadMatBigram: Row %d of bigram %s adds up to %f",1,fn,sum);
for (p=2; ReadLMWord(buf); p++) {
if (trace&T_TIO) {
if ((p%25)==0)
printf(". "),fflush(stdout);
if ((p%800)==0)
printf("\n "),fflush(stdout);
}
if (p>P)
HError(8150,"ReadMatBigram: More rows than columns in bigram %s",fn);
id=GetLabId(buf,TRUE);
if ((int)id->aux != 0)
HError(8150,"ReadMatBigram: Duplicated name %s in bigram %s",buf,fn);
id->aux = (Ptr) p;
matbi->wdlist[p] = id;
SkipWhiteSpace(&source);
if (ReadRow(matbi->bigMat[p])!=P)
HError(8150,"ReadMatBigram: Wrong number of items in row %d",p);
for (sum=0.0, j=1; j<=P; j++) {
x = matbi->bigMat[p][j];
if (x<0)
HError(8151,"ReadMatBigram: In bigram, entry %d for %s is -ve (%e)",
j,buf,x);
sum += x;
matbi->bigMat[p][j]=((x<MINLARG)?LZERO:log(x));
}
if (sum < 0.99 || sum > 1.01)
HError(-8151,"ReadMatBigram: Row %d of bigram %s adds up to %f",p,fn,sum);
}
if (P>p)
HError(8150,"ReadMatBigram: More columns than rows in bigram %s",fn);
if (trace&T_TIO)
printf("\n"),fflush(stdout);
CloseSource(&source);
}
