1POCKETSPHINX_CONTINUOUS(1)  General Commands Manual POCKETSPHINX_CONTINUOUS(1)
2
3
4

NAME

6       pocketsphinx_continuous  - Run speech recognition in continuous listen‐
7       ing mode
8

SYNOPSIS

10       pocketsphinx_continuous -hmm hmmdir -dict dictfile [ options ]...
11

DESCRIPTION

13       This program opens the audio device and  waits  for  speech.   When  it
14       detects an utterance, it performs speech recognition on it.
15
16       -adcdev
17              name for audio input (platform-specific)
18
19       -adchdr
20              Size of audio file header in bytes (headers are ignored)
21
22       -adcin Input is raw audio data
23
24       -agc   Automatic  gain  control  for  c0  ('max',  'emax',  'noise', or
25              'none')
26
27       -agcthresh
28              Initial threshold for automatic gain control
29
30       -allphone
31              Do phoneme recognition
32
33       -alpha Preemphasis parameter
34
35       -backtrace
36              Print back trace of recognition results
37
38       -beam  Beam width applied to every frame  in  Viterbi  search  (smaller
39              values mean wider beam)
40
41       -bestpath
42              Run bestpath (Dijkstra) search over word lattice (3rd pass)
43
44       -bestpathlw
45              Language model probability weight for bestpath search
46
47       -cachesen
48              Cache senone scores from first pass search
49
50       -cep2spec
51              Input is cepstral files, output is log spectral files
52
53       -cepdir
54              files directory (prefixed to filespecs in control file)
55
56       -cepext
57              Input files extension (prefixed to filespecs in control file)
58
59       -ceplen
60              Number of components in the input feature vector
61
62       -cmn   Cepstral  mean  normalization  scheme  ('current',  'prior',  or
63              'none')
64
65       -cmninit
66              Initial values (comma-separated) for cepstral mean when  'prior'
67              is used
68
69       -compallsen
70              Compute  all  senone  scores  in every frame (can be faster when
71              there are many senones)
72
73       -ctl   file listing utterances to be processed
74
75       -ctlcount
76              No. of utterances to be  processed  (after  skipping  -ctloffset
77              entries)
78
79       -ctlincr
80              Do every Nth line in the control file
81
82       -ctloffset
83              No. of utterances at the beginning of -ctl file to be skipped
84
85       -dict  pronunciation dictionary (lexicon) input file
86
87       -dither
88              Add 1/2-bit noise
89
90       -doublebw
91              Use double bandwidth filters (same center freq)
92
93       -dsratio
94              Frame GMM computation downsampling ratio
95
96       -fbtype
97              FB Type of mel_scale or log_linear
98
99       -fdict word pronunciation dictionary input file
100
101       -feat  Feature stream type, depends on the acoustic model
102
103       -fillpen
104              Filler word transition penalty
105
106       -frate Frame rate
107
108       -fsg   state grammar
109
110       -fsgbfs
111              Force backtrace from FSG final state
112
113       -fsgctlfn
114              finite state grammar control file
115
116       -fsgusealtpron
117              Use alternative pronunciations for FSG
118
119       -fsgusefiller
120              (FSG Mode (Mode 2) only) Insert filler words at each state.
121
122       -fwd3g Use trigrams in first pass search
123
124       -fwdflat
125              Run forward flat-lexicon search over word lattice (2nd pass)
126
127       -fwdflatbeam
128              Beam width applied to every frame in second-pass flat search
129
130       -fwdflatefwid
131              Minimum  number  of end frames for a word to be searched in fwd‐
132              flat search
133
134       -fwdflatlw
135              Language model probability weight for flat  lexicon  (2nd  pass)
136              decoding
137
138       -fwdflatsfwin
139              Window  of  frames  in  lattice to search for successor words in
140              fwdflat search
141
142       -fwdflatwbeam
143              Beam width applied to word exits in second-pass flat search
144
145       -fwdtree
146              Run forward lexicon-tree search (1st pass)
147
148       -hmm   containing acoustic model files.
149
150       -hyp   output file name
151
152       -hypseg
153              output with segmentation file name
154
155       -input_endian
156              Endianness of input data, big or little, ignored if NIST  or  MS
157              Wav
158
159       -kdmaxbbi
160              Maximum number of Gaussians per leaf node in kd-Trees
161
162       -kdmaxdepth
163              Maximum depth of kd-Trees to use
164
165       -kdtree
166              file for Gaussian selection
167
168       -latsize
169              Lattice size
170
171       -lifter
172              Length of sin-curve for liftering, or 0 for no liftering.
173
174       -live  Get input from audio hardware
175
176       -lm    trigram language model input file
177
178       -lmctl a set of language model
179
180       The  -hmm  and -dict arguments are always required.  Either -lm or -fsg
181       is required, depending on whether you are using a statistical  language
182       model or a finite-state grammar.
183

AUTHOR

185       Written  by numerous people at CMU from 1994 onwards.  This manual page
186       by David Huggins-Daines <dhuggins@cs.cmu.edu>
187

COPYRIGHT

189       Copyright © 1994-2007 Carnegie Mellon University.  See the file COPYING
190       included with this package for more information.
191

SEE ALSO

193       pocketsphinx_batch(1), sphinx_fe(1).
194
195
196
197                                  2007-08-27        POCKETSPHINX_CONTINUOUS(1)