static fsg_model_t *
jsgf_build_fsg_internal(jsgf_t * grammar, jsgf_rule_t * rule,
logmath_t * lmath, float32 lw, int do_closure)
{
fsg_model_t *fsg;
glist_t nulls;
gnode_t *gn;
int rule_entry, rule_exit;
/* Clear previous links */
for (gn = grammar->links; gn; gn = gnode_next(gn)) {
ckd_free(gnode_ptr(gn));
}
glist_free(grammar->links);
grammar->links = NULL;
grammar->nstate = 0;
/* Create the top-level entry state, and expand the
top-level rule. */
rule_entry = grammar->nstate++;
rule_exit = expand_rule(grammar, rule, rule_entry, NO_NODE);
/* If no exit-state was created, create one. */
if (rule_exit == NO_NODE) {
rule_exit = grammar->nstate++;
jsgf_add_link(grammar, NULL, rule_entry, rule_exit);
}
fsg = fsg_model_init(rule->name, lmath, lw, grammar->nstate);
fsg->start_state = rule_entry;
fsg->final_state = rule_exit;
grammar->links = glist_reverse(grammar->links);
for (gn = grammar->links; gn; gn = gnode_next(gn)) {
jsgf_link_t *link = gnode_ptr(gn);
if (link->atom) {
if (jsgf_atom_is_rule(link->atom)) {
fsg_model_null_trans_add(fsg, link->from, link->to,
logmath_log(lmath,
link->atom->weight));
}
else {
int wid = fsg_model_word_add(fsg, link->atom->name);
fsg_model_trans_add(fsg, link->from, link->to,
logmath_log(lmath, link->atom->weight),
wid);
}
}
else {
fsg_model_null_trans_add(fsg, link->from, link->to, 0);
}
}
if (do_closure) {
nulls = fsg_model_null_trans_closure(fsg, NULL);
glist_free(nulls);
}
return fsg;
}
int
main(int argc, char *argv[])
{
jsgf_t *jsgf;
fsg_model_t *fsg;
cmd_ln_t *config;
const char *rule;
if ((config = cmd_ln_parse_r(NULL, defn, argc, argv, TRUE)) == NULL)
return 1;
if (cmd_ln_boolean_r(config, "-help")) {
usagemsg(argv[0]);
}
jsgf = jsgf_parse_file(cmd_ln_str_r(config, "-jsgf"), NULL);
if (jsgf == NULL) {
return 1;
}
rule = cmd_ln_str_r(config, "-toprule") ? cmd_ln_str_r(config, "-toprule") : NULL;
if (!(fsg = get_fsg(jsgf, rule))) {
E_ERROR("No fsg was built for the given rule '%s'.\n"
"Check rule name; it should be qualified (with grammar name)\n"
"and not enclosed in angle brackets (e.g. 'grammar.rulename').",
rule);
return 1;
}
if (cmd_ln_boolean_r(config, "-compile")) {
fsg_model_null_trans_closure(fsg, NULL);
}
if (cmd_ln_str_r(config, "-fsm")) {
const char* outfile = cmd_ln_str_r(config, "-fsm");
const char* symfile = cmd_ln_str_r(config, "-symtab");
if (outfile)
fsg_model_writefile_fsm(fsg, outfile);
else
fsg_model_write_fsm(fsg, stdout);
if (symfile)
fsg_model_writefile_symtab(fsg, symfile);
}
else {
const char *outfile = cmd_ln_str_r(config, "-fsg");
if (outfile)
fsg_model_writefile(fsg, outfile);
else
fsg_model_write(fsg, stdout);
}
fsg_model_free(fsg);
jsgf_grammar_free(jsgf);
return 0;
}
int
main(int argc, char *argv[])
{
jsgf_t *jsgf;
fsg_model_t *fsg;
cmd_ln_t *config;
if ((config = cmd_ln_parse_r(NULL, defn, argc, argv, TRUE)) == NULL)
return 1;
if (cmd_ln_boolean_r(config, "-help")) {
usagemsg(argv[0]);
}
jsgf = jsgf_parse_file(cmd_ln_str_r(config, "-jsgf"), NULL);
if (jsgf == NULL) {
return 1;
}
fsg = get_fsg(jsgf, cmd_ln_str_r(config, "-rule") ? cmd_ln_str_r(config, "-rule") : NULL);
if (cmd_ln_boolean_r(config, "-compile")) {
fsg_model_null_trans_closure(fsg, NULL);
}
if (cmd_ln_str_r(config, "-fsm")) {
const char* outfile = cmd_ln_str_r(config, "-fsm");
const char* symfile = cmd_ln_str_r(config, "-symtab");
if (outfile)
fsg_model_writefile_fsm(fsg, outfile);
else
fsg_model_write_fsm(fsg, stdout);
if (symfile)
fsg_model_writefile_symtab(fsg, symfile);
}
else {
const char *outfile = cmd_ln_str_r(config, "-fsg");
if (outfile)
fsg_model_writefile(fsg, outfile);
else
fsg_model_write(fsg, stdout);
}
fsg_model_free(fsg);
jsgf_grammar_free(jsgf);
return 0;
}
static fsg_model_t *
jsgf_build_fsg_internal(jsgf_t *grammar, jsgf_rule_t *rule,
logmath_t *lmath, float32 lw, int do_closure)
{
fsg_model_t *fsg;
glist_t nulls;
gnode_t *gn;
/* Clear previous links */
for (gn = grammar->links; gn; gn = gnode_next(gn)) {
ckd_free(gnode_ptr(gn));
}
glist_free(grammar->links);
grammar->links = NULL;
rule->entry = rule->exit = 0;
grammar->nstate = 0;
expand_rule(grammar, rule);
fsg = fsg_model_init(rule->name, lmath, lw, grammar->nstate);
fsg->start_state = rule->entry;
fsg->final_state = rule->exit;
grammar->links = glist_reverse(grammar->links);
for (gn = grammar->links; gn; gn = gnode_next(gn)) {
jsgf_link_t *link = gnode_ptr(gn);
if (link->atom) {
if (jsgf_atom_is_rule(link->atom)) {
fsg_model_null_trans_add(fsg, link->from, link->to,
logmath_log(lmath, link->atom->weight));
}
else {
int wid = fsg_model_word_add(fsg, link->atom->name);
fsg_model_trans_add(fsg, link->from, link->to,
logmath_log(lmath, link->atom->weight), wid);
}
}
else {
fsg_model_null_trans_add(fsg, link->from, link->to, 0);
}
}
if (do_closure) {
nulls = fsg_model_null_trans_closure(fsg, NULL);
glist_free(nulls);
}
return fsg;
}
fsg_model_t *
fsg_model_read(FILE * fp, logmath_t * lmath, float32 lw)
{
fsg_model_t *fsg;
hash_table_t *vocab;
hash_iter_t *itor;
int32 lastwid;
char **wordptr;
char *lineptr;
char *fsgname;
int32 lineno;
int32 n, i, j;
int n_state, n_trans, n_null_trans;
glist_t nulls;
float32 p;
lineno = 0;
vocab = hash_table_new(32, FALSE);
wordptr = NULL;
lineptr = NULL;
nulls = NULL;
fsgname = NULL;
fsg = NULL;
/* Scan upto FSG_BEGIN header */
for (;;) {
n = nextline_str2words(fp, &lineno, &lineptr, &wordptr);
if (n < 0) {
E_ERROR("%s declaration missing\n", FSG_MODEL_BEGIN_DECL);
goto parse_error;
}
if ((strcmp(wordptr[0], FSG_MODEL_BEGIN_DECL) == 0)) {
if (n > 2) {
E_ERROR("Line[%d]: malformed FSG_BEGIN declaration\n",
lineno);
goto parse_error;
}
break;
}
}
/* Save FSG name, or it will get clobbered below :(.
* If name is missing, try the default.
*/
if (n == 2) {
fsgname = ckd_salloc(wordptr[1]);
}
else {
E_WARN("FSG name is missing\n");
fsgname = ckd_salloc("unknown");
}
/* Read #states */
n = nextline_str2words(fp, &lineno, &lineptr, &wordptr);
if ((n != 2)
|| ((strcmp(wordptr[0], FSG_MODEL_N_DECL) != 0)
&& (strcmp(wordptr[0], FSG_MODEL_NUM_STATES_DECL) != 0))
|| (sscanf(wordptr[1], "%d", &n_state) != 1)
|| (n_state <= 0)) {
E_ERROR
("Line[%d]: #states declaration line missing or malformed\n",
lineno);
goto parse_error;
}
/* Now create the FSG. */
fsg = fsg_model_init(fsgname, lmath, lw, n_state);
ckd_free(fsgname);
fsgname = NULL;
/* Read start state */
n = nextline_str2words(fp, &lineno, &lineptr, &wordptr);
if ((n != 2)
|| ((strcmp(wordptr[0], FSG_MODEL_S_DECL) != 0)
&& (strcmp(wordptr[0], FSG_MODEL_START_STATE_DECL) != 0))
|| (sscanf(wordptr[1], "%d", &(fsg->start_state)) != 1)
|| (fsg->start_state < 0)
|| (fsg->start_state >= fsg->n_state)) {
E_ERROR
("Line[%d]: start state declaration line missing or malformed\n",
lineno);
goto parse_error;
}
/* Read final state */
n = nextline_str2words(fp, &lineno, &lineptr, &wordptr);
if ((n != 2)
|| ((strcmp(wordptr[0], FSG_MODEL_F_DECL) != 0)
&& (strcmp(wordptr[0], FSG_MODEL_FINAL_STATE_DECL) != 0))
|| (sscanf(wordptr[1], "%d", &(fsg->final_state)) != 1)
|| (fsg->final_state < 0)
|| (fsg->final_state >= fsg->n_state)) {
E_ERROR
("Line[%d]: final state declaration line missing or malformed\n",
lineno);
goto parse_error;
}
/* Read transitions */
lastwid = 0;
n_trans = n_null_trans = 0;
for (;;) {
int32 wid, tprob;
n = nextline_str2words(fp, &lineno, &lineptr, &wordptr);
if (n <= 0) {
E_ERROR("Line[%d]: transition or FSG_END statement expected\n",
lineno);
goto parse_error;
}
if ((strcmp(wordptr[0], FSG_MODEL_END_DECL) == 0)) {
break;
}
if ((strcmp(wordptr[0], FSG_MODEL_T_DECL) == 0)
|| (strcmp(wordptr[0], FSG_MODEL_TRANSITION_DECL) == 0)) {
if (((n != 4) && (n != 5))
|| (sscanf(wordptr[1], "%d", &i) != 1)
|| (sscanf(wordptr[2], "%d", &j) != 1)
|| (i < 0) || (i >= fsg->n_state)
|| (j < 0) || (j >= fsg->n_state)) {
E_ERROR
("Line[%d]: transition spec malformed; Expecting: from-state to-state trans-prob [word]\n",
lineno);
goto parse_error;
}
p = atof_c(wordptr[3]);
if ((p <= 0.0) || (p > 1.0)) {
E_ERROR
("Line[%d]: transition spec malformed; Expecting float as transition probability\n",
lineno);
goto parse_error;
}
}
else {
E_ERROR("Line[%d]: transition or FSG_END statement expected\n",
lineno);
goto parse_error;
}
tprob = (int32) (logmath_log(lmath, p) * fsg->lw);
/* Add word to "dictionary". */
if (n > 4) {
if (hash_table_lookup_int32(vocab, wordptr[4], &wid) < 0) {
(void) hash_table_enter_int32(vocab,
ckd_salloc(wordptr[4]),
lastwid);
wid = lastwid;
++lastwid;
}
fsg_model_trans_add(fsg, i, j, tprob, wid);
++n_trans;
}
else {
if (fsg_model_null_trans_add(fsg, i, j, tprob) == 1) {
++n_null_trans;
nulls =
glist_add_ptr(nulls, fsg_model_null_trans(fsg, i, j));
}
}
}
E_INFO("FSG: %d states, %d unique words, %d transitions (%d null)\n",
fsg->n_state, hash_table_inuse(vocab), n_trans, n_null_trans);
/* Now create a string table from the "dictionary" */
fsg->n_word = hash_table_inuse(vocab);
fsg->n_word_alloc = fsg->n_word + 10; /* Pad it a bit. */
fsg->vocab = ckd_calloc(fsg->n_word_alloc, sizeof(*fsg->vocab));
for (itor = hash_table_iter(vocab); itor;
itor = hash_table_iter_next(itor)) {
char const *word = hash_entry_key(itor->ent);
int32 wid = (int32) (long) hash_entry_val(itor->ent);
fsg->vocab[wid] = (char *) word;
}
hash_table_free(vocab);
/* Do transitive closure on null transitions */
nulls = fsg_model_null_trans_closure(fsg, nulls);
glist_free(nulls);
ckd_free(lineptr);
ckd_free(wordptr);
return fsg;
parse_error:
for (itor = hash_table_iter(vocab); itor;
itor = hash_table_iter_next(itor))
ckd_free((char *) hash_entry_key(itor->ent));
glist_free(nulls);
hash_table_free(vocab);
ckd_free(fsgname);
ckd_free(lineptr);
ckd_free(wordptr);
fsg_model_free(fsg);
return NULL;
}
