int
mdef_convert_main(int argc, char *argv[])
{
const char *infile, *outfile;
bin_mdef_t *bin;
int tobin = 1;
if (argc < 3 || argc > 4) {
fprintf(stderr, "Usage: %s [-text | -bin] INPUT OUTPUT\n",
argv[0]);
return 1;
}
if (argv[1][0] == '-') {
if (strcmp(argv[1], "-text") == 0) {
tobin = 0;
++argv;
}
else if (strcmp(argv[1], "-bin") == 0) {
tobin = 1;
++argv;
}
else {
fprintf(stderr, "Unknown argument %s\n", argv[1]);
fprintf(stderr, "Usage: %s [-text | -bin] INPUT OUTPUT\n",
argv[0]);
return 1;
}
}
infile = argv[1];
outfile = argv[2];
if (tobin) {
if ((bin = bin_mdef_read_text(NULL, infile)) == NULL) {
fprintf(stderr, "Failed to read text mdef from %s\n", infile);
return 1;
}
if (bin_mdef_write(bin, outfile) < 0) {
fprintf(stderr, "Failed to write binary mdef to %s\n",
outfile);
return 1;
}
}
else {
if ((bin = bin_mdef_read(NULL, infile)) == NULL) {
fprintf(stderr, "Failed to read binary mdef from %s\n",
infile);
return 1;
}
if (bin_mdef_write_text(bin, outfile) < 0) {
fprintf(stderr, "Failed to write text mdef to %s\n", outfile);
return 1;
}
}
return 0;
}
int
main(int argc, char *argv[])
{
bin_mdef_t *mdef;
dict_t *dict;
cmd_ln_t *config;
int i;
char buf[100];
TEST_ASSERT(config = cmd_ln_init(NULL, NULL, FALSE,
"-dict", MODELDIR "/en-us/cmudict-en-us.dict",
"-fdict", MODELDIR "/en-us/en-us/noisedict",
NULL));
/* Test dictionary in standard fashion. */
TEST_ASSERT(mdef = bin_mdef_read(NULL, MODELDIR "/en-us/en-us/mdef"));
TEST_ASSERT(dict = dict_init(config, mdef, NULL));
printf("Word ID (CARNEGIE) = %d\n",
dict_wordid(dict, "CARNEGIE"));
printf("Word ID (ASDFASFASSD) = %d\n",
dict_wordid(dict, "ASDFASFASSD"));
TEST_EQUAL(0, dict_write(dict, "_cmu07a.dic", NULL));
TEST_EQUAL(0, system("diff -uw " MODELDIR "/en-us/cmudict-en-us.dict _cmu07a.dic"));
dict_free(dict);
bin_mdef_free(mdef);
/* Now test an empty dictionary. */
TEST_ASSERT(dict = dict_init(NULL, NULL, NULL));
printf("Word ID(<s>) = %d\n", dict_wordid(dict, "<s>"));
TEST_ASSERT(BAD_S3WID != dict_add_word(dict, "FOOBIE", NULL, 0));
TEST_ASSERT(BAD_S3WID != dict_add_word(dict, "BLETCH", NULL, 0));
printf("Word ID(FOOBIE) = %d\n", dict_wordid(dict, "FOOBIE"));
printf("Word ID(BLETCH) = %d\n", dict_wordid(dict, "BLETCH"));
TEST_ASSERT(dict_real_word(dict, dict_wordid(dict, "FOOBIE")));
TEST_ASSERT(dict_real_word(dict, dict_wordid(dict, "BLETCH")));
TEST_ASSERT(!dict_real_word(dict, dict_wordid(dict, "</s>")));
dict_free(dict);
/* Test to add 500k words. */
TEST_ASSERT(dict = dict_init(NULL, NULL, NULL));
for (i = 0; i < 500000; i++) {
sprintf(buf, "word_%d", i);
TEST_ASSERT(BAD_S3WID != dict_add_word(dict, buf, NULL, 0));
}
dict_free(dict);
cmd_ln_free_r(config);
return 0;
}
int
main(int argc, char *argv[])
{
ngram_trie_t *t;
dict_t *dict;
bin_mdef_t *mdef;
logmath_t *lmath;
cmd_ln_t *config;
FILE *arpafh;
config = cmd_ln_init(NULL, ps_args(), TRUE,
"-hmm", TESTDATADIR "/hub4wsj_sc_8k",
"-dict", TESTDATADIR "/bn10000.homos.dic",
NULL);
ps_init_defaults(config);
lmath = logmath_init(cmd_ln_float32_r(config, "-logbase"),
0, FALSE);
mdef = bin_mdef_read(config, cmd_ln_str_r(config, "-mdef"));
dict = dict_init(config, mdef);
t = ngram_trie_init(dict, lmath);
arpafh = fopen(TESTDATADIR "/bn10000.3g.arpa", "r");
ngram_trie_read_arpa(t, arpafh);
fclose(arpafh);
/* Test 1, 2, 3-gram probs without backoff. */
test_lookups(t, lmath);
arpafh = fopen("tmp.bn10000.3g.arpa", "w");
ngram_trie_write_arpa(t, arpafh);
fclose(arpafh);
ngram_trie_free(t);
t = ngram_trie_init(dict, lmath);
arpafh = fopen("tmp.bn10000.3g.arpa", "r");
ngram_trie_read_arpa(t, arpafh);
fclose(arpafh);
/* Test 1, 2, 3-gram probs without backoff. */
test_lookups(t, lmath);
/* Test adding nodes. */
test_add_nodes(t, lmath);
ngram_trie_free(t);
dict_free(dict);
logmath_free(lmath);
bin_mdef_free(mdef);
cmd_ln_free_r(config);
return 0;
}
static int
acmod_init_am(acmod_t *acmod)
{
char const *mdeffn, *tmatfn;
/* Read model definition. */
if ((mdeffn = cmd_ln_str_r(acmod->config, "-mdef")) == NULL) {
E_ERROR("Must specify -mdef or -hmm\n");
return -1;
}
if ((acmod->mdef = bin_mdef_read(acmod->config, mdeffn)) == NULL) {
E_ERROR("Failed to read model definition from %s\n", mdeffn);
return -1;
}
/* Read transition matrices. */
if ((tmatfn = cmd_ln_str_r(acmod->config, "-tmat")) == NULL) {
E_ERROR("No tmat file specified\n");
return -1;
}
acmod->tmat = tmat_init(tmatfn, acmod->lmath,
cmd_ln_float32_r(acmod->config, "-tmatfloor"),
TRUE);
/* Read the acoustic models. */
if ((cmd_ln_str_r(acmod->config, "-mean") == NULL)
|| (cmd_ln_str_r(acmod->config, "-var") == NULL)
|| (cmd_ln_str_r(acmod->config, "-tmat") == NULL)) {
E_ERROR("No mean/var/tmat files specified\n");
return -1;
}
if (cmd_ln_str_r(acmod->config, "-senmgau")) {
E_INFO("Using general multi-stream GMM computation\n");
acmod->mgau = ms_mgau_init(acmod->config, acmod->lmath, acmod->mdef);
if (acmod->mgau == NULL)
return -1;
}
else {
E_INFO("Attempting to use SCHMM computation module\n");
if ((acmod->mgau = s2_semi_mgau_init(acmod)) == NULL) {
E_INFO("Attempting to use PTHMM computation module\n");
if ((acmod->mgau = ptm_mgau_init(acmod)) == NULL) {
E_INFO("Falling back to general multi-stream GMM computation\n");
acmod->mgau = ms_mgau_init(acmod->config, acmod->lmath, acmod->mdef);
if (acmod->mgau == NULL)
return -1;
}
}
}
return 0;
}
int
main(int argc, char *argv[])
{
bin_mdef_t *mdef;
dict_t *dict;
dict2pid_t *d2p;
cmd_ln_t *config;
TEST_ASSERT(config = cmd_ln_init(NULL, NULL, FALSE,
"-dict", MODELDIR "/lm/en_US/cmu07a.dic",
"-fdict", MODELDIR "/hmm/en_US/hub4wsj_sc_8k/noisedict",
NULL));
TEST_ASSERT(mdef = bin_mdef_read(NULL, MODELDIR "/hmm/en_US/hub4wsj_sc_8k/mdef"));
TEST_ASSERT(dict = dict_init(config, mdef));
TEST_ASSERT(d2p = dict2pid_build(mdef, dict));
dict_free(dict);
dict2pid_free(d2p);
bin_mdef_free(mdef);
cmd_ln_free_r(config);
return 0;
}
int
main(int argc, char *argv[])
{
dict_t *dict;
dict2pid_t *d2p;
bin_mdef_t *mdef;
glextree_t *tree;
hmm_context_t *ctx;
logmath_t *lmath;
tmat_t *tmat;
int i;
TEST_ASSERT(mdef = bin_mdef_read(NULL, MODELDIR "/hmm/en_US/hub4wsj_sc_8k/mdef"));
TEST_ASSERT(dict = dict_init(cmd_ln_init(NULL, NULL, FALSE,
"-dict", DATADIR "/turtle.dic",
"-dictcase", "no", NULL),
mdef));
TEST_ASSERT(d2p = dict2pid_build(mdef, dict));
lmath = logmath_init(1.0001, 0, TRUE);
tmat = tmat_init(MODELDIR "/hmm/en_US/hub4wsj_sc_8k/transition_matrices",
lmath, 1e-5, TRUE);
ctx = hmm_context_init(bin_mdef_n_emit_state(mdef), tmat->tp, NULL, mdef->sseq);
TEST_ASSERT(tree = glextree_build(ctx, dict, d2p, NULL, NULL));
/* Check that a path exists for all dictionary words. */
for (i = 0; i < dict_size(dict); ++i)
TEST_ASSERT(glextree_has_word(tree, i));
dict_free(dict);
dict2pid_free(d2p);
bin_mdef_free(mdef);
tmat_free(tmat);
hmm_context_free(ctx);
glextree_free(tree);
return 0;
}
static int
acmod_init_am(acmod_t *acmod)
{
char const *mdeffn, *tmatfn, *mllrfn, *hmmdir;
/* Read model definition. */
if ((mdeffn = cmd_ln_str_r(acmod->config, "-mdef")) == NULL) {
if ((hmmdir = cmd_ln_str_r(acmod->config, "-hmm")) == NULL)
E_ERROR("Acoustic model definition is not specified either "
"with -mdef option or with -hmm\n");
else
E_ERROR("Folder '%s' does not contain acoustic model "
"definition 'mdef'\n", hmmdir);
return -1;
}
if ((acmod->mdef = bin_mdef_read(acmod->config, mdeffn)) == NULL) {
E_ERROR("Failed to read acoustic model definition from %s\n", mdeffn);
return -1;
}
/* Read transition matrices. */
if ((tmatfn = cmd_ln_str_r(acmod->config, "-tmat")) == NULL) {
E_ERROR("No tmat file specified\n");
return -1;
}
acmod->tmat = tmat_init(tmatfn, acmod->lmath,
cmd_ln_float32_r(acmod->config, "-tmatfloor"),
TRUE);
/* Read the acoustic models. */
if ((cmd_ln_str_r(acmod->config, "-mean") == NULL)
|| (cmd_ln_str_r(acmod->config, "-var") == NULL)
|| (cmd_ln_str_r(acmod->config, "-tmat") == NULL)) {
E_ERROR("No mean/var/tmat files specified\n");
return -1;
}
if (cmd_ln_str_r(acmod->config, "-senmgau")) {
E_INFO("Using general multi-stream GMM computation\n");
acmod->mgau = ms_mgau_init(acmod, acmod->lmath, acmod->mdef);
if (acmod->mgau == NULL)
return -1;
}
else {
E_INFO("Attempting to use PTM computation module\n");
if ((acmod->mgau = ptm_mgau_init(acmod, acmod->mdef)) == NULL) {
E_INFO("Attempting to use semi-continuous computation module\n");
if ((acmod->mgau = s2_semi_mgau_init(acmod)) == NULL) {
E_INFO("Falling back to general multi-stream GMM computation\n");
acmod->mgau = ms_mgau_init(acmod, acmod->lmath, acmod->mdef);
if (acmod->mgau == NULL)
return -1;
}
}
}
/* If there is an MLLR transform, apply it. */
if ((mllrfn = cmd_ln_str_r(acmod->config, "-mllr"))) {
ps_mllr_t *mllr = ps_mllr_read(mllrfn);
if (mllr == NULL)
return -1;
acmod_update_mllr(acmod, mllr);
}
return 0;
}
