1POCKETSPHINX_BATCH(1) General Commands Manual POCKETSPHINX_BATCH(1)
2
6 pocketsphinx_batch - Run speech recognition in batch mode
7
9 pocketsphinx_batch -ctl ctlfile -cepdir cepdir -cepext .mfc [ options
10 ]...
11
13 Run speech recognition over a list of utterances in batchmode. A list
14 of arguments follows:
15 -adchdr
17 Size of audio file header in bytes (headers are ignored)
18 -adcin Input is raw audio data
20 -agc Automatic gain control for c0 ('max', 'emax', 'noise', or
22 'none')
23 -agcthresh
25 Initial threshold for automatic gain control
26 -allphone
28 phoneme decoding with phonetic lm
29 -allphone_ci
31 Perform phoneme decoding with phonetic lm and context-indepen‐
32 dent units only
33 -alpha Preemphasis parameter
35 -argfile
37 file giving extra arguments.
38 -ascale
40 Inverse of acoustic model scale for confidence score calculation
41 -aw Inverse weight applied to acoustic scores.
43 -backtrace
45 Print results and backtraces to log file.
46 -beam Beam width applied to every frame in Viterbi search (smaller
48 values mean wider beam)
49 -bestpath
51 Run bestpath (Dijkstra) search over word lattice (3rd pass)
52 -bestpathlw
54 Language model probability weight for bestpath search
55 -build_outdirs
57 Create missing subdirectories in output directory
58 -cepdir
60 files directory (prefixed to filespecs in control file)
61 -cepext
63 Input files extension (suffixed to filespecs in control file)
64 -ceplen
66 Number of components in the input feature vector
67 -cmn Cepstral mean normalization scheme ('current', 'prior', or
69 'none')
70 -cmninit
72 Initial values (comma-separated) for cepstral mean when 'prior'
73 is used
74 -compallsen
76 Compute all senone scores in every frame (can be faster when
77 there are many senones)
78 -ctl file listing utterances to be processed
80 -ctlcount
82 No. of utterances to be processed (after skipping -ctloffset
83 entries)
84 -ctlincr
86 Do every Nth line in the control file
87 -ctloffset
89 No. of utterances at the beginning of -ctl file to be skipped
90 -ctm output in CTM file format (may require post-sorting)
92 -debug level for debugging messages
94 -dict pronunciation dictionary (lexicon) input file
96 -dictcase
98 Dictionary is case sensitive (NOTE: case insensitivity applies
99 to ASCII characters only)
100 -dither
102 Add 1/2-bit noise
103 -doublebw
105 Use double bandwidth filters (same center freq)
106 -ds Frame GMM computation downsampling ratio
108 -fdict word pronunciation dictionary input file
110 -feat Feature stream type, depends on the acoustic model
112 -featparams
114 containing feature extraction parameters.
115 -fillprob
117 Filler word transition probability
118 -frate Frame rate
120 -fsg format finite state grammar file
122 -fsgctl
124 file listing FSG file to use for each utterance
125 -fsgdir
127 directory for FSG files
128 -fsgext
130 extension for FSG files (including leading dot)
131 -fsgusealtpron
133 Add alternate pronunciations to FSG
134 -fsgusefiller
136 Insert filler words at each state.
137 -fwdflat
139 Run forward flat-lexicon search over word lattice (2nd pass)
140 -fwdflatbeam
142 Beam width applied to every frame in second-pass flat search
143 -fwdflatefwid
145 Minimum number of end frames for a word to be searched in fwd‐
146 flat search
147 -fwdflatlw
149 Language model probability weight for flat lexicon (2nd pass)
150 decoding
151 -fwdflatsfwin
153 Window of frames in lattice to search for successor words in
154 fwdflat search
155 -fwdflatwbeam
157 Beam width applied to word exits in second-pass flat search
158 -fwdtree
160 Run forward lexicon-tree search (1st pass)
161 -hmm containing acoustic model files.
163 -hyp output file name
165 -hypseg
167 output with segmentation file name
168 -input_endian
170 Endianness of input data, big or little, ignored if NIST or MS
171 Wav
172 -jsgf grammar file
174 -keyphrase
176 to spot
177 -kws file with keyphrases to spot, one per line
179 -kws_delay
181 Delay to wait for best detection score
182 -kws_plp
184 Phone loop probability for keyword spotting
185 -kws_threshold
187 Threshold for p(hyp)/p(alternatives) ratio
188 -latsize
190 Initial backpointer table size
191 -lda containing transformation matrix to be applied to features (sin‐
193 gle-stream features only)
194 -ldadim
196 Dimensionality of output of feature transformation (0 to use
197 entire matrix)
198 -lifter
200 Length of sin-curve for liftering, or 0 for no liftering.
201 -lm trigram language model input file
203 -lmctl a set of language model
205 -lmname
207 language model in -lmctl to use by default
208 -lmnamectl
210 file listing LM name to use for each utterance
211 -logbase
213 Base in which all log-likelihoods calculated
214 -logfn to write log messages in
216 -logspec
218 Write out logspectral files instead of cepstra
219 -lowerf
221 Lower edge of filters
222 -lpbeam
224 Beam width applied to last phone in words
225 -lponlybeam
227 Beam width applied to last phone in single-phone words
228 -lw Language model probability weight
230 -maxhmmpf
232 Maximum number of active HMMs to maintain at each frame (or -1
233 for no pruning)
234 -maxwpf
236 Maximum number of distinct word exits at each frame (or -1 for
237 no pruning)
238 -mdef definition input file
240 -mean gaussian means input file
242 -mfclogdir
244 to log feature files to
245 -min_endfr
247 Nodes ignored in lattice construction if they persist for fewer
248 than N frames
249 -mixw mixture weights input file (uncompressed)
251 -mixwfloor
253 Senone mixture weights floor (applied to data from -mixw file)
254 -mllr transformation to apply to means and variances
256 -mllrctl
258 file listing MLLR transforms to use for each utterance
259 -mllrdir
261 directory for MLLR transforms
262 -mllrext
264 extension for MLLR transforms (including leading dot)
265 -mmap Use memory-mapped I/O (if possible) for model files
267 -nbest Number of N-best hypotheses to write to -nbestdir (0 for no N-
269 best)
270 -nbestdir
272 for writing N-best hypothesis lists
273 -nbestext
275 Extension for N-best hypothesis list files
276 -ncep Number of cep coefficients
278 -nfft Size of FFT
280 -nfilt Number of filter banks
282 -nwpen New word transition penalty
284 -outlatbeam
286 Minimum posterior probability for output lattice nodes
287 -outlatdir
289 for dumping word lattices
290 -outlatext
292 Filename extension for dumping word lattices
293 -outlatfmt
295 Format for dumping word lattices (s3 or htk)
296 -pbeam Beam width applied to phone transitions
298 -pip Phone insertion penalty
300 -pl_beam
302 Beam width applied to phone loop search for lookahead
303 -pl_pbeam
305 Beam width applied to phone loop transitions for lookahead
306 -pl_pip
308 Phone insertion penalty for phone loop
309 -pl_weight
311 Weight for phoneme lookahead penalties
312 -pl_window
314 Phoneme lookahead window size, in frames
315 -rawlogdir
317 to log raw audio files to
318 -remove_dc
320 Remove DC offset from each frame
321 -remove_noise
323 Remove noise with spectral subtraction in mel-energies
324 -remove_silence
326 Enables VAD, removes silence frames from processing
327 -round_filters
329 Round mel filter frequencies to DFT points
330 -samprate
332 Sampling rate
333 -seed Seed for random number generator; if less than zero, pick our
335 own
336 -sendump
338 dump (compressed mixture weights) input file
339 -senin Input is senone score dump files
341 -senlogdir
343 to log senone score files to
344 -senmgau
346 to codebook mapping input file (usually not needed)
347 -silprob
349 Silence word transition probability
350 -smoothspec
352 Write out cepstral-smoothed logspectral files
353 -svspec
355 specification (e.g., 24,0-11/25,12-23/26-38 or 0-12/13-25/26-38)
356 -tmat state transition matrix input file
358 -tmatfloor
360 HMM state transition probability floor (applied to -tmat file)
361 -topn Maximum number of top Gaussians to use in scoring.
363 -topn_beam
365 Beam width used to determine top-N Gaussians (or a list, per-
366 feature)
367 -toprule
369 rule for JSGF (first public rule is default)
370 -transform
372 Which type of transform to use to calculate cepstra (legacy,
373 dct, or htk)
374 -unit_area
376 Normalize mel filters to unit area
377 -upperf
379 Upper edge of filters
380 -uw Unigram weight
382 -vad_postspeech
384 Num of silence frames to keep after from speech to silence.
385 -vad_prespeech
387 Num of speech frames to keep before silence to speech.
388 -vad_startspeech
390 Num of speech frames to trigger vad from silence to speech.
391 -vad_threshold
393 Threshold for decision between noise and silence frames. Log-
394 ratio between signal level and noise level.
395 -var gaussian variances input file
397 -varfloor
399 Mixture gaussian variance floor (applied to data from -var file)
400 -varnorm
402 Variance normalize each utterance (only if CMN == current)
403 -verbose
405 Show input filenames
406 -warp_params
408 defining the warping function
409 -warp_type
411 Warping function type (or shape)
412 -wbeam Beam width applied to word exits
414 -wip Word insertion penalty
416 -wlen Hamming window length
418 To do batchmode recognition, you will need to specify a control file,
420 using -ctl This is a simple text file containing one entry per line.
421 Each entry is the name of an input file relative to the -cepdir direc‐
422 tory, and without the filename extension (which is given in the -cepext
423 argument).
424 If you are using acoustic feature files as input (see sphinx_fe(1) for
426 information on how to generate these), you can also specify a subpart
427 of a file, using the following format:
428 FILENAME START-FRAME END-FRAME UTTERANCE-ID
430
432 Written by numerous people at CMU from 1994 onwards. This manual page
433 by David Huggins-Daines <dhuggins@cs.cmu.edu>
434
436 Copyright © 1994-2016 Carnegie Mellon University. See the file LICENSE
437 included with this package for more information.
438
440 pocketsphinx_continuous(1), sphinx_fe(1).
441 2007-08-27 POCKETSPHINX_BATCH(1)