1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
|
#include <torch/LinearLexicon.h>
#include <torch/LanguageModel.h>
#include <torch/DecoderBatchTest.h>
#include <torch/BeamSearchDecoder.h>
#include <torch/Vocabulary.h>
#include <torch/DiskXFile.h>
#include <torch/CmdLine.h>
#include <torch/PhoneInfo.h>
#include <torch/PhoneModels.h>
#include <torch/LexiconInfo.h>
using namespace Torch ;
// Acoustic Modelling Parameters
bool am_input_vecs_are_ftrs=false ;
char *am_models_fname=NULL ;
char *am_sil_phone=NULL ;
char *am_pause_phone=NULL ;
real am_phone_del_pen=1.0 ;
bool am_apply_pause_del_pen=false ;
char *am_priors_fname=NULL ;
char *am_mlp_fname=NULL ;
int am_mlp_cw_size=9 ;
char *am_norms_fname=NULL ;
bool am_online_norm_ftrs=false ;
real am_online_norm_alpha_m=0.005 ;
real am_online_norm_alpha_v=0.005 ;
// Lexicon Parameters
char *lex_dict_fname=NULL ;
char *lex_sent_start_word=NULL ;
char *lex_sent_end_word=NULL ;
char *lex_sil_word=NULL ;
// Language Model Parameters
int lm_ngram_order=0 ;
char *lm_fname=NULL ;
real lm_scaling_factor=1.0 ;
// Beam Search Decoder Parameters
real dec_int_prune_window=LOG_ZERO ;
real dec_end_prune_window=LOG_ZERO ;
real dec_word_entr_pen=0.0 ;
bool dec_verbose=false ;
bool dec_delayed_lm=true ;
// Batch Test Parameters
char *input_fname=NULL ;
char *input_format_s=NULL ;
DSTDataFileFormat input_format=DST_PROBS_LNA8BIT ;
char *wrdtrns_fname=NULL ;
char *output_fname=NULL ;
bool output_ctm=false ;
real msec_step_size=10.0 ;
void processCmdLine( CmdLine *cmd , int argc , char *argv[] )
{
// Phoneset Options
cmd->addText("\nPhoneset Options:") ;
cmd->addSCmdOption( "-am_models_fname" , &am_models_fname , "" ,
"the file with the HMM definitions for the phone models" ) ;
cmd->addSCmdOption( "-am_sil_phone" , &am_sil_phone , "" ,
"the name of silence phoneme" ) ;
cmd->addSCmdOption( "-am_pause_phone" , &am_pause_phone , "" ,
"the name of pause phoneme" ) ;
cmd->addRCmdOption( "-am_phone_del_pen" , &am_phone_del_pen , 1.0 ,
"the (non-log) phone deletion penalty" ) ;
cmd->addBCmdOption( "-am_apply_pause_del_pen" , &am_apply_pause_del_pen , false ,
"indicates whether the phone deletion penalty is applied to pause phone" ) ;
cmd->addSCmdOption( "-am_priors_fname" , &am_priors_fname , "" ,
"the (priors format) file containing the phone prior probabilities" ) ;
cmd->addSCmdOption( "-am_mlp_fname" , &am_mlp_fname , "" ,
"the file (MLPW binary format) containing MLP weights" ) ;
cmd->addICmdOption( "-am_mlp_cw_size" , &am_mlp_cw_size , 9 ,
"the number of frames in the context window input to the MLP" ) ;
cmd->addSCmdOption( "-am_norms_fname" , &am_norms_fname , "" ,
"the (norms format) file with means & inv stddevs for ftr normalisation") ;
cmd->addBCmdOption( "-am_online_norm_ftrs" , &am_online_norm_ftrs , false ,
"indicates whether online normalisation of features is to be performed") ;
cmd->addRCmdOption( "-am_online_norm_alpha_m" , &am_online_norm_alpha_m , 0.005 ,
"parameter used to control adaptation of feature means" ) ;
cmd->addRCmdOption( "-am_online_norm_alpha_v" , &am_online_norm_alpha_v , 0.005 ,
"parameter used to control adaptation of feature variances" ) ;
// Lexicon Parameters
cmd->addText("\nLexicon Options:") ;
cmd->addSCmdOption( "-lex_dict_fname" , &lex_dict_fname , "" ,
"the dictionary file" ) ;
cmd->addSCmdOption( "-lex_sent_start_word" , &lex_sent_start_word , "" ,
"the name of the dictionary word that will start every sentence" ) ;
cmd->addSCmdOption( "-lex_sent_end_word" , &lex_sent_end_word , "" ,
"the name of the dictionary word that will end every sentence" ) ;
cmd->addSCmdOption( "-lex_sil_word" , &lex_sil_word , "" ,
"the name of the silence dictionary word" ) ;
// Language Model Parameters
cmd->addText("\nLanguage Model Options:") ;
cmd->addSCmdOption( "-lm_fname" , &lm_fname , "" ,
"the file (ARPA LM format) containing the LM probabilities" ) ;
cmd->addICmdOption( "-lm_ngram_order" , &lm_ngram_order , 0 ,
"the order of the n-gram language model" ) ;
cmd->addRCmdOption( "-lm_scaling_factor" , &lm_scaling_factor , 1.0 ,
"the factor by which log LM probs are scaled during decoding" ) ;
// Beam Search Decoder Parameters
cmd->addText("\nBeam Search Decoding Options:") ;
cmd->addRCmdOption( "-dec_int_prune_window" , &dec_int_prune_window , LOG_ZERO ,
"the (+ve log) window used for pruning word-interior state hypotheses" ) ;
cmd->addRCmdOption( "-dec_end_prune_window" , &dec_end_prune_window , LOG_ZERO ,
"the (+ve log) window used for pruning word-end state hypotheses" ) ;
cmd->addRCmdOption( "-dec_word_entr_pen" , &dec_word_entr_pen , 0.0 ,
"the log word entrance penalty" ) ;
cmd->addBCmdOption( "-dec_delayed_lm" , &dec_delayed_lm , false ,
"indicates whether LM probabilities are applied in a delayed fashion" ) ;
cmd->addBCmdOption( "-dec_verbose" , &dec_verbose , false ,
"indicates whether frame-by-frame decoding info is printed (to stderr)" ) ;
// General Parameters
cmd->addText("\nGeneral Options:") ;
cmd->addSCmdOption( "-input_fname" , &input_fname , "" ,
"the file containing the list of files to be decoded (or an archive)" ) ;
cmd->addSCmdOption( "-input_format" , &input_format_s , "" ,
"the format of the input files (htk,lna,lna_archive,online_ftrs,online_ftrs_archive)" ) ;
cmd->addSCmdOption( "-output_fname" , &output_fname , "" ,
"the file where decoding results are written" ) ;
cmd->addSCmdOption( "-wrdtrns_fname" , &wrdtrns_fname , "" ,
"the file containing word-level reference transcriptions" ) ;
cmd->addBCmdOption( "-output_ctm" , &output_ctm , false ,
"indicates whether output is to be written in CTM format" ) ;
cmd->addRCmdOption( "-msec_step_size" , &msec_step_size , 10.0 ,
"the frame step size in msec used with CTM output" ) ;
cmd->read( argc , argv ) ;
// First interpret the input_format
if ( strcmp( input_format_s , "" ) == 0 )
error("input_format undefined\n") ;
if ( strcmp( input_format_s , "htk" ) == 0 )
{
am_input_vecs_are_ftrs = true ;
input_format = DST_FEATS_HTK ;
}
else if ( strcmp( input_format_s , "lna" ) == 0 )
{
am_input_vecs_are_ftrs = false ;
input_format = DST_PROBS_LNA8BIT ;
}
else if ( strcmp( input_format_s , "lna_archive" ) == 0 )
{
am_input_vecs_are_ftrs = false ;
input_format = DST_PROBS_LNA8BIT_ARCHIVE ;
}
else if ( strcmp( input_format_s , "online_ftrs" ) == 0 )
{
am_input_vecs_are_ftrs = true ;
input_format = DST_FEATS_ONLINE_FTRS ;
}
else if ( strcmp( input_format_s , "online_ftrs_archive" ) == 0 )
{
am_input_vecs_are_ftrs = true ;
input_format = DST_FEATS_ONLINE_FTRS_ARCHIVE ;
}
else
error("-input_format %s : unrecognised format\n",input_format_s) ;
// Basic parameter checks
if ( strcmp( input_fname , "" ) == 0 )
error("input_fname undefined\n") ;
if ( strcmp( am_models_fname , "" ) == 0 )
error("am_models_fname undefined\n") ;
if ( strcmp( lex_dict_fname , "" ) == 0 )
error("lex_dict_fname undefined\n") ;
// Some 2 parameter dependencies
if ( (strcmp(am_mlp_fname,"") != 0) && (am_input_vecs_are_ftrs == false) )
error("am_mlp_fname specified but input vectors are not features\n") ;
if ( (lm_ngram_order > 0) && (strcmp(lm_fname,"") == 0) )
error("lm_ngram_order > 0 but no LM file specified\n") ;
}
int main( int argc , char *argv[] )
{
CmdLine cmd ;
DiskXFile::setBigEndianMode() ;
processCmdLine( &cmd , argc , argv ) ;
LexiconInfo lex_info( am_models_fname , am_sil_phone , am_pause_phone , lex_dict_fname ,
lex_sent_start_word , lex_sent_end_word , lex_sil_word ) ;
PhoneModels phone_models ( lex_info.phone_info , am_models_fname ,
am_input_vecs_are_ftrs , am_phone_del_pen ,
am_apply_pause_del_pen , am_priors_fname , am_mlp_fname ,
am_mlp_cw_size , am_norms_fname , am_online_norm_ftrs ,
am_online_norm_alpha_m , am_online_norm_alpha_v ) ;
LinearLexicon lexicon( &lex_info , &phone_models ) ;
LanguageModel *lang_model ;
if ( lm_ngram_order <= 0 )
lang_model = NULL ;
else
{
lang_model = new LanguageModel( lm_ngram_order , lex_info.vocabulary ,
lm_fname , lm_scaling_factor ) ;
}
BeamSearchDecoder bs_decoder( &lexicon , lang_model , dec_word_entr_pen ,
dec_int_prune_window , dec_end_prune_window ,
dec_delayed_lm , dec_verbose ) ;
DecoderBatchTest batch_tester( input_fname , input_format , wrdtrns_fname , &bs_decoder ,
true , true , output_fname , output_ctm , msec_step_size ) ;
batch_tester.run() ;
if ( lang_model != NULL )
delete lang_model ;
return(0) ;
}
|
