/**
* Load and initialize a word list for isolated word recognition.
*
* @param winfo [out] pointer to a word dictionary data to store the read data
* @param filename [in] file name of the word dictionary to read
* @param hmminfo [in] %HMM definition data, needed for triphone conversion.
* @param headphone [in] word head silence phone name
* @param tailphone [in] word tail silence phone name
* @param conextphone [in] silence context name at head and tail phoneme
* @param force_dict [in] TRUE if want to ignore the error words in the dictionary
*
* @return TRUE on success, FALSE on failure.
*/
boolean
init_wordlist(WORD_INFO *winfo, char *filename, HTK_HMM_INFO *hmminfo, char *headphone, char *tailphone, char *contextphone, boolean force_dict)
{
FILE *fd;
jlog("Stat: init_wordlist: reading in word list\n");
if ((fd = fopen_readfile(filename)) == NULL) {
jlog("Error: init_wordlist: failed to open %s\n",filename);
return(FALSE);
}
if (!voca_load_wordlist(fd, winfo, hmminfo, headphone, tailphone, contextphone)) {
if (force_dict) {
jlog("Warning: init_wordlist: the word errors are ignored\n");
} else {
jlog("Error: init_wordlist: error in reading %s: %d words failed out of %d words\n",filename, winfo->errnum, winfo->num);
fclose_readfile(fd);
return(FALSE);
}
}
if (fclose_readfile(fd) == -1) {
jlog("Error: init_wordlist: failed to close\n");
return(FALSE);
}
jlog("Stat: init_wordlist: read %d words\n", winfo->num);
return(TRUE);
}
/**
* Load and initialize a word dictionary.
*
* @param winfo [out] pointer to a word dictionary data to store the read data
* @param filename [in] file name of the word dictionary to read
* @param hmminfo [in] %HMM definition data, needed for triphone conversion.
* @param not_conv_tri [in] TRUE if not converting monophone to triphone.
* @param force_dict [in] TRUE if want to ignore the error words in the dictionary
*
* @return TRUE on success, FALSE on failure.
*/
boolean
init_voca(WORD_INFO *winfo, char *filename, HTK_HMM_INFO *hmminfo, boolean not_conv_tri, boolean force_dict)
{
FILE *fd;
if ((fd = fopen_readfile(filename)) == NULL) {
jlog("Error: init_voca: failed to open %s\n",filename);
return(FALSE);
}
if (!voca_load_htkdict(fd, winfo, hmminfo, not_conv_tri)) {
if (force_dict) {
jlog("Warning: init_voca: the word errors are ignored\n");
} else {
jlog("Error: init_voca: error in reading %s: %d words failed out of %d words\n",filename, winfo->errnum, winfo->num);
fclose_readfile(fd);
return(FALSE);
}
}
if (fclose_readfile(fd) == -1) {
jlog("Error: init_voca: failed to close\n");
return(FALSE);
}
jlog("Stat: init_voca: read %d words\n", winfo->num);
return(TRUE);
}
void
init_term(char *filename, char **termname)
{
FILE *fd;
int n;
static char buf[512];
char *p;
fprintf(stderr, "Reading in term file (optional)...");
if ((fd = fopen_readfile(filename)) == NULL) {
termname[0] = NULL;
fprintf(stderr, "not found\n");
return;
}
while (getl(buf, sizeof(buf), fd) != NULL) {
if ((p = strtok(buf, DELM)) == NULL) {
fprintf(stderr, "Error: term file failed to parse, corrupted or invalid data?\n");
return;
}
n = atoi(p);
if ((p = strtok(NULL, DELM)) == NULL) {
fprintf(stderr, "Error: term file failed to parse, corrupted or invalid data?\n");
return;
}
termname[n] = strdup(p);
}
if (fclose_readfile(fd) == -1) {
fprintf(stderr, "close error\n");
exit(1);
}
fprintf(stderr, "done\n");
}
/**
* Load a noise spectrum from file.
*
* @param filename [in] path name of noise spectrum file
* @param slen [out] length of the returned buffer
*
* @return a newly allocated buffer that holds the loaded noise spectrum.
*/
float *
new_SS_load_from_file(char *filename, int *slen)
{
FILE *fp;
int num;
float *sbuf;
/* open file */
jlog("Stat: ss: reading Noise Spectrum for SS\n");
if ((fp = fopen_readfile(filename)) == NULL) {
jlog("Error: ss: failed to open \"%s\"\n", filename);
return(NULL);
}
/* read length */
if (myread(&num, sizeof(int), 1, fp) == FALSE) {
jlog("Error: ss: failed to read \"%s\"\n", filename);
return(NULL);
}
/* allocate */
sbuf = (float *)mymalloc(sizeof(float) * num);
/* read data */
if (myread(sbuf, sizeof(float), num, fp) == FALSE) {
jlog("Error: ss: failed to read \"%s\"\n", filename);
return(NULL);
}
/* close file */
fclose_readfile(fp);
*slen = num;
jlog("Stat: ss: done\n");
return(sbuf);
}
/**
* Top function to read a HTK parameter file.
*
* @param filename [in] HTK parameter file name
* @param pinfo [in] parameter data (already allocated by new_param())
*
* @return TRUE on success, FALSE on failure.
*/
boolean
rdparam(char *filename, HTK_Param *pinfo)
{
FILE *fp;
boolean retflag;
if ((fp = fopen_readfile(filename)) == NULL) return(FALSE);
retflag = read_param(fp, pinfo);
if (fclose_readfile(fp) < 0) return (FALSE);
return (retflag);
}
/**
* Read in a grammar file and set to DFA grammar structure
*
* @param dinfo [i/o] a blank DFA data
* @param filename [in] DFA grammar file name
*/
boolean
init_dfa(DFA_INFO *dinfo, char *filename)
{
FILE *fd;
if ((fd = fopen_readfile(filename)) == NULL) {
jlog("Error: init_dfa: failed to open %s\n",filename);
return FALSE;
}
if (!rddfa(fd, dinfo)) {
jlog("Error; init_dfa: error in reading %s\n",filename);
return FALSE;
}
if (fclose_readfile(fd) == -1) {
jlog("Error: init_dfa: failed to close %s\n", filename);
return FALSE;
}
return TRUE;
}
/**
* Load CMN parameter from file. If the number of MFCC dimension in the
* file does not match the specified one, an error will occur.
*
* Format can be either HTK ascii format or binary format (made by Julius older than ver.4.2.3)
*
* @param c [i/o] CMN calculation work area
* @param filename [in] file name
*
* @return TRUE on success, FALSE on failure.
*/
boolean
CMN_load_from_file(CMNWork *c, char *filename)
{
FILE *fp;
int veclen;
char ch[5];
char *buf;
jlog("Stat: wav2mfcc-pipe: reading initial cepstral mean/variance from file \"%s\"\n", filename);
if ((fp = fopen_readfile(filename)) == NULL) {
jlog("Error: wav2mfcc-pipe: failed to open %s\n", filename);
return(FALSE);
}
/* detect file format */
if (myread(&ch, sizeof(char), 5, fp) == FALSE) {
jlog("Error: wav2mfcc-pipe: failed to read CMN/CVN file\n");
fclose_readfile(fp);
return(FALSE);
}
myfrewind(fp);
if (ch[0] == '<' &&
(ch[1] == 'C' || ch[1] == 'c') &&
(ch[2] == 'E' || ch[2] == 'e') &&
(ch[3] == 'P' || ch[3] == 'p') &&
(ch[4] == 'S' || ch[4] == 's') ) {
/* ascii HTK format (>=4.3) */
char *p;
int mode;
int d, dv, len;
jlog("Stat: wav2mfcc-pipe: reading HTK-format cepstral vectors\n");
buf = (char *)mymalloc(MAXBUFLEN);
mode = 0;
while(getl(buf, MAXBUFLEN, fp) != NULL) {
for (p = mystrtok_quote(buf, "<> \t\r\n"); p; p = mystrtok_quote(NULL, "<> \t\r\n")) {
switch(mode){
case 0:
if (strmatch(p, "MEAN")) {
mode = 1;
} else if (strmatch(p, "VARIANCE")) {
mode = 3;
}
break;
case 1:
len = atof(p);
if (len != c->veclen && len != c->mfcc_dim) {
jlog("Error: wav2mfcc-pipe: cepstral dimension mismatch\n");
jlog("Error: wav2mfcc-pipe: process = %d (%d), file = %d\n", c->veclen, c->mfcc_dim, len);
free(buf); fclose_readfile(fp);
return(FALSE);
}
for (d = 0; d < c->veclen; d++) c->cmean_init[d] = 0.0;
d = 0;
mode = 2;
break;
case 2:
if (strmatch(p, "VARIANCE")) {
mode = 3;
} else {
if (d >= len) {
jlog("Error: wav2mfcc-pipe: corrupted data\n");
free(buf); fclose_readfile(fp);
return(FALSE);
}
c->cmean_init[d++] = atof(p);
}
break;
case 3:
len = atof(p);
if (len != c->veclen) {
jlog("Error: wav2mfcc-pipe: cepstral dimension mismatch\n");
jlog("Error: wav2mfcc-pipe: process = %d, file = %d\n", c->veclen, len);
free(buf); fclose_readfile(fp);
return(FALSE);
}
dv = 0;
mode = 4;
break;
case 4:
if (dv >= len) {
jlog("Error: wav2mfcc-pipe: corrupted data\n");
free(buf); fclose_readfile(fp);
return(FALSE);
}
c->cvar_init[dv++] = atof(p);
break;
}
}
}
free(buf);
if (d != len || (mode >= 3 && dv != len)) {
jlog("Error: wav2mfcc-pipe: corrupted data\n");
fclose_readfile(fp);
return(FALSE);
}
} else {
/* binary (<4.3) */
jlog("Stat: wav2mfcc-pipe: reading binary-format cepstral vectors\n");
/* read header */
if (myread(&veclen, sizeof(int), 1, fp) == FALSE) {
jlog("Error: wav2mfcc-pipe: failed to read header\n");
fclose_readfile(fp);
return(FALSE);
}
/* check length */
if (veclen != c->veclen) {
jlog("Error: wav2mfcc-pipe: cepstral dimension mismatch\n");
jlog("Error: wav2mfcc-pipe: process = %d, file = %d\n", c->veclen, veclen);
fclose_readfile(fp);
return(FALSE);
}
/* read body */
if (myread(c->cmean_init, sizeof(float), c->veclen, fp) == FALSE) {
jlog("Error: wav2mfcc-pipe: failed to read mean for CMN\n");
fclose_readfile(fp);
return(FALSE);
}
if (c->var) {
if (myread(c->cvar_init, sizeof(float), c->veclen, fp) == FALSE) {
jlog("Error: wav2mfcc-pipe: failed to read variance for CVN\n");
fclose_readfile(fp);
return(FALSE);
}
}
}
if (fclose_readfile(fp) == -1) {
jlog("Error: wav2mfcc-pipe: failed to close\n");
return(FALSE);
}
c->cmean_init_set = TRUE;
c->loaded_from_file = TRUE;
jlog("Stat: wav2mfcc-pipe: finished reading CMN/CVN parameter\n");
return(TRUE);
}