-
Notifications
You must be signed in to change notification settings - Fork 2
/
clara3.jconf
191 lines (163 loc) · 8.1 KB
/
clara3.jconf
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
######################################################################
# Julius Conf File
######################################################################
######################################################################
### GRAMMAR
######################################################################
-dfa clara.dfa -v clara.dict
######################################################################
### Speech segment detection by level and zero-cross
######################################################################
-cutsilence # detection on.
#-nocutsilence # detection off.
-lv 3000 # level threshold (0-32767).
-zc 150 # zero-cross threshold (times in sec.).
-headmargin 400 # head silence margin (msec).
-tailmargin 400 # tail silence margin (msec).
-rejectshort 250 # reject shorter input (msec).
######################################################################
### Decoding option
######################################################################
-realtime # force real-time processing.
#-norealtime # force non real-time processin.
######################################################################
### Speech Input Source (Audio Input)
######################################################################
## select one (default: mfcfile)
#-input mfcfile # MFCC file in HTK parameter file format
#-input rawfile # raw wavefile (auto-detect format)
# WAV(16bit) or
# RAW(16bit(signed short),mono,big-endian)
# AIFF,AU (with libsndfile extension)
# other than 16kHz, sampling rate should
# be specified by "-smpFreq" option
-input mic # direct microphone input
# device name can be specified via env.
# val. "AUDIODEV"
#-input netaudio -NA host:0 # direct input from DatLink(NetAudio) host
#-input adinnet -adport portnum # via adinnet network client
#-input stdin # from standard tty input (pipe)
#-filelist audiolist # specify file list to be recognized in batch mode.
#-notypecheck # will disab. the check and use the input vector as is.
#-nostrip # switch OFF dropping of invalid input segment.
-zmean # enable/disabled DC offset removal
#-nozmean # (invalid for mfcfile input)
# (default: disabled)
######################################################################
### Isolated Word
######################################################################
## Multiple dictionary
#-w clara.dict # word dictionary.
#-wsil sil sil NULL # head / tail silence models to be appended.
######################################################################
#### Speech Analysis Parameters
######################################################################
#-smpPeriod 227 # sampling period (ns) (= 10000000 / smpFreq)
# (default: 625 = 16000Hz).
-smpFreq 16000 # sampling rate (Hz)
# (default: 16000).
#-fsize 400 # window size (samples)
# (default: 400).
#-fshift 160 # frame shift (samples)
# (default: 160).
#-delwin 2 # delta window (frames)
# (default: 2).
#-hifreq -1 # cut-off hi frequency (Hz) (-1: disable)
# (default: -1).
#-lofreq -1 # cut-off low frequency (Hz) (-1: disable)
# (default: -1).
#-preemph 0.97 # Pre-emphasis coefficient.
# (default: 0.97).
#-fbank 24 # Number of filterbank channels.
# (default:24).
-enormal # Enable/disable normalizing log energy
#-noenormal # (default: disabled Julius / enable HTK).
-zmeanframe # with speech input enables/disables
#-nozmeanframe # frame-wise DC offset removal
# (default: disabled).
######################################################################
#### Real-time Cepstral Mean Normalization
######################################################################
#-cmnload filename # load initial CMN param from file on .
#-cmnsave filename # save CMN param to file (update per input, overwrite).
######################################################################
#### Misc AM Options
######################################################################
-htkconf edaz.conf # parameter values are switched from Julius defaults
# to HTK defaults.
######################################################################
#### Recognizer and Search ( -SR )
######################################################################
## Context-dependency handling will be enabled according to the model type.
## Try below if julius wrongly detect the type of hmmdefs.
#-no_ccd # disable context-dependency handling.
-force_ccd # enable context-dependency handling.
-cmalpha 0.05 # Smoothing patemeter for confidence scoring
# (default: 0.05).
## For insertion of context-free short-term inter-word pauses between words
## (multi-path version only)
-iwsp # append a skippable sp model at all word ends
-transp -20.0 # additional insertion penalty for transparent words.
# (default: 0.0)
######################################################################
#### 1st Pass Parameters
######################################################################
-lmp 15.0 10.0 # (weight penalty) LM weights and words insertion
# penalties for 1st pass. (N-gram)
-penalty1 -5.0 # (Grammar) word insertion penalty for the first pass
# (default: 0.0).
-b 2000 # Envelope beam width (number of hypothesis) for
# rank beam in number of HMM nodes on the 1st pass
# (decoder become very slow)... 400 (monophone),
# 800 (triphone), or 1000 (triphone, setup=v2.1).
-progout # Enable progressive output of the partial results
# on the 1st pass.
######################################################################
#### 2st Pass Parameters
######################################################################
-lmp2 15.0 10.0 # (weight penalty) LM weights and words insertion
# penalties for 2st pass. (N-gram)
-b2 200 # Envelope beam width (number of hypothesis) for
# rank beam in number of HMM nodes on the 2st pass
# (decoder become very slow) (default: 30).
-sb 300 # Score envelope width for enveloped scoring.
# (default: 80.0)
#-s 500 # Stack size i.e. hypotheses stack size on 2nd pass.
# (default: 500)
#-m 10000 # hypotheses overflow threshold
# (default: 2000).
#-n 1 # number of candidates Julius tries to find
# (default: 10 (standard) or 1 (fast or v2.1)). X
#-output 1 # top N sentence hypo. to be output at the end of search
# (default: 1).
#-lookuprange 5 # lookup range for word expansion
# (default: 5).
#-looktrellis # words survived on the first pass instead of expanding
# all the words predicted by grammar
# (deafult: disabled).
######################################################################
### Acoustic HMM and Parameters
######################################################################
-h hmmdefs # HMM model(ascii format or
# Julius binary format).
-hlist tiedlist # triphone model needs HMMList that maps
# logical triphone to physical ones.
#-tmix 2 # Gaussian Pruning, number of mixtures to
# select in a mixture pdf. This default value
# is optimized for IPA99's PTM, with 64
# Gaussians per codebook.
-spmodel "sp" # Short pause model name to be treated
# as special HMM model norealtime.
## Select Gaussian pruning algorithm.
## default: safe (standard), beam (fast).
-gprune safe # safe pruning, accurate but slow.
#-gprune heuristic # heuristic pruning.
#-gprune beam # beam pruning, fast but sensitive.
#-gprune none # no pruning.
## (PTM/triphone) switch computation method of IWCD on 1st pass.
## default: -iwcd1 avg
#-iwcd1 best 5 # assign average of N-best likelihood
# of the same context.
#-iwcd1 max # assign maximum likelihood of the same context.
#-iwcd1 avg # assign average likelihood of the same context.
#-iwsppenalty -5.0 # transition penalty for the appended sp models.