1POCKETSPHINX_CONTINUOUS(1) General Commands Manual POCKETSPHINX_CONTINUOUS(1)
2
6 pocketsphinx_continuous - Run speech recognition in continuous listen‐
7 ing mode
8
10 pocketsphinx_continuous [-infile filename.wav ] [ -inmic yes ] [
11 options ]...
12
14 This program opens the audio device or a file and waits for speech.
15 When it detects an utterance, it performs speech recognition on it.
16 To record from microphone and decode use
18 -inmic yes
20 To decode a 16kHz 16-bit mono WAV file use
22 -infile filename.wav
24 You can also specify -lm or -fsg or -kws depending on whether you are
26 using a statistical language model or a finite-state grammar or look
27 for a keyphase.
28
30 -adcdev
31 of audio device to use for input.
32 -agc Automatic gain control for c0 ('max', 'emax', 'noise', or
34 'none')
35 -agcthresh
37 Initial threshold for automatic gain control
38 -allphone
40 phoneme decoding with phonetic lm
41 -allphone_ci
43 Perform phoneme decoding with phonetic lm and context-indepen‐
44 dent units only
45 -alpha Preemphasis parameter
47 -argfile
49 file giving extra arguments.
50 -ascale
52 Inverse of acoustic model scale for confidence score calculation
53 -aw Inverse weight applied to acoustic scores.
55 -backtrace
57 Print results and backtraces to log file.
58 -beam Beam width applied to every frame in Viterbi search (smaller
60 values mean wider beam)
61 -bestpath
63 Run bestpath (Dijkstra) search over word lattice (3rd pass)
64 -bestpathlw
66 Language model probability weight for bestpath search
67 -ceplen
69 Number of components in the input feature vector
70 -cmn Cepstral mean normalization scheme ('current', 'prior', or
72 'none')
73 -cmninit
75 Initial values (comma-separated) for cepstral mean when 'prior'
76 is used
77 -compallsen
79 Compute all senone scores in every frame (can be faster when
80 there are many senones)
81 -debug level for debugging messages
83 -dict pronunciation dictionary (lexicon) input file
85 -dictcase
87 Dictionary is case sensitive (NOTE: case insensitivity applies
88 to ASCII characters only)
89 -dither
91 Add 1/2-bit noise
92 -doublebw
94 Use double bandwidth filters (same center freq)
95 -ds Frame GMM computation downsampling ratio
97 -fdict word pronunciation dictionary input file
99 -feat Feature stream type, depends on the acoustic model
101 -featparams
103 containing feature extraction parameters.
104 -fillprob
106 Filler word transition probability
107 -frate Frame rate
109 -fsg format finite state grammar file
111 -fsgusealtpron
113 Add alternate pronunciations to FSG
114 -fsgusefiller
116 Insert filler words at each state.
117 -fwdflat
119 Run forward flat-lexicon search over word lattice (2nd pass)
120 -fwdflatbeam
122 Beam width applied to every frame in second-pass flat search
123 -fwdflatefwid
125 Minimum number of end frames for a word to be searched in fwd‐
126 flat search
127 -fwdflatlw
129 Language model probability weight for flat lexicon (2nd pass)
130 decoding
131 -fwdflatsfwin
133 Window of frames in lattice to search for successor words in
134 fwdflat search
135 -fwdflatwbeam
137 Beam width applied to word exits in second-pass flat search
138 -fwdtree
140 Run forward lexicon-tree search (1st pass)
141 -hmm containing acoustic model files.
143 -infile
145 file to transcribe.
146 -inmic Transcribe audio from microphone.
148 -input_endian
150 Endianness of input data, big or little, ignored if NIST or MS
151 Wav
152 -jsgf grammar file
154 -keyphrase
156 to spot
157 -kws file with keyphrases to spot, one per line
159 -kws_delay
161 Delay to wait for best detection score
162 -kws_plp
164 Phone loop probability for keyword spotting
165 -kws_threshold
167 Threshold for p(hyp)/p(alternatives) ratio
168 -latsize
170 Initial backpointer table size
171 -lda containing transformation matrix to be applied to features (sin‐
173 gle-stream features only)
174 -ldadim
176 Dimensionality of output of feature transformation (0 to use
177 entire matrix)
178 -lifter
180 Length of sin-curve for liftering, or 0 for no liftering.
181 -lm trigram language model input file
183 -lmctl a set of language model
185 -lmname
187 language model in -lmctl to use by default
188 -logbase
190 Base in which all log-likelihoods calculated
191 -logfn to write log messages in
193 -logspec
195 Write out logspectral files instead of cepstra
196 -lowerf
198 Lower edge of filters
199 -lpbeam
201 Beam width applied to last phone in words
202 -lponlybeam
204 Beam width applied to last phone in single-phone words
205 -lw Language model probability weight
207 -maxhmmpf
209 Maximum number of active HMMs to maintain at each frame (or -1
210 for no pruning)
211 -maxwpf
213 Maximum number of distinct word exits at each frame (or -1 for
214 no pruning)
215 -mdef definition input file
217 -mean gaussian means input file
219 -mfclogdir
221 to log feature files to
222 -min_endfr
224 Nodes ignored in lattice construction if they persist for fewer
225 than N frames
226 -mixw mixture weights input file (uncompressed)
228 -mixwfloor
230 Senone mixture weights floor (applied to data from -mixw file)
231 -mllr transformation to apply to means and variances
233 -mmap Use memory-mapped I/O (if possible) for model files
235 -ncep Number of cep coefficients
237 -nfft Size of FFT
239 -nfilt Number of filter banks
241 -nwpen New word transition penalty
243 -pbeam Beam width applied to phone transitions
245 -pip Phone insertion penalty
247 -pl_beam
249 Beam width applied to phone loop search for lookahead
250 -pl_pbeam
252 Beam width applied to phone loop transitions for lookahead
253 -pl_pip
255 Phone insertion penalty for phone loop
256 -pl_weight
258 Weight for phoneme lookahead penalties
259 -pl_window
261 Phoneme lookahead window size, in frames
262 -rawlogdir
264 to log raw audio files to
265 -remove_dc
267 Remove DC offset from each frame
268 -remove_noise
270 Remove noise with spectral subtraction in mel-energies
271 -remove_silence
273 Enables VAD, removes silence frames from processing
274 -round_filters
276 Round mel filter frequencies to DFT points
277 -samprate
279 Sampling rate
280 -seed Seed for random number generator; if less than zero, pick our
282 own
283 -sendump
285 dump (compressed mixture weights) input file
286 -senlogdir
288 to log senone score files to
289 -senmgau
291 to codebook mapping input file (usually not needed)
292 -silprob
294 Silence word transition probability
295 -smoothspec
297 Write out cepstral-smoothed logspectral files
298 -svspec
300 specification (e.g., 24,0-11/25,12-23/26-38 or 0-12/13-25/26-38)
301 -time Print word times in file transcription.
303 -tmat state transition matrix input file
305 -tmatfloor
307 HMM state transition probability floor (applied to -tmat file)
308 -topn Maximum number of top Gaussians to use in scoring.
310 -topn_beam
312 Beam width used to determine top-N Gaussians (or a list, per-
313 feature)
314 -toprule
316 rule for JSGF (first public rule is default)
317 -transform
319 Which type of transform to use to calculate cepstra (legacy,
320 dct, or htk)
321 -unit_area
323 Normalize mel filters to unit area
324 -upperf
326 Upper edge of filters
327 -uw Unigram weight
329 -vad_postspeech
331 Num of silence frames to keep after from speech to silence.
332 -vad_prespeech
334 Num of speech frames to keep before silence to speech.
335 -vad_startspeech
337 Num of speech frames to trigger vad from silence to speech.
338 -vad_threshold
340 Threshold for decision between noise and silence frames. Log-
341 ratio between signal level and noise level.
342 -var gaussian variances input file
344 -varfloor
346 Mixture gaussian variance floor (applied to data from -var file)
347 -varnorm
349 Variance normalize each utterance (only if CMN == current)
350 -verbose
352 Show input filenames
353 -warp_params
355 defining the warping function
356 -warp_type
358 Warping function type (or shape)
359 -wbeam Beam width applied to word exits
361 -wip Word insertion penalty
363 -wlen Hamming window length
365
367 Written by numerous people at CMU from 1994 onwards. This manual page
368 by David Huggins-Daines <dhuggins@cs.cmu.edu>
369
371 Copyright © 1994-2016 Carnegie Mellon University. See the file LICENSE
372 included with this package for more information.
373
375 pocketsphinx_batch(1), sphinx_fe(1).
376 2016-04-01 POCKETSPHINX_CONTINUOUS(1)