| 9 | | {{{ |
| 10 | | # Copyright (c) 1991-2005 Kawahara Lab., Kyoto University |
| 11 | | # Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and Technology |
| 12 | | # Copyright (c) 2005 Julius project team, Nagoya Institute of Technology |
| 13 | | # All rights reserved |
| 14 | | # |
| 15 | | # Example of Runtime Configuration File for Julian rev.3.5 |
| 16 | | # |
| 17 | | # 1) NOTICE: relative paths must be relative to THIS FILE. |
| 18 | | # Current working directory does not affect. |
| 19 | | # 2) Blank lines and comments after '#' are ignored. |
| 20 | | # Use `\#' to specify `#' in an argument, |
| 21 | | # 3) Each line should be shorter than 512 bytes. |
| 22 | | # 4) Syntax of each arguments are same as corresponding command options. |
| 23 | | # 5) The commented-out values in this file are default values in Julian. |
| 24 | | # |
| 25 | | |
| 26 | | ###################################################################### |
| 27 | | #### Files |
| 28 | | ###################################################################### |
| 29 | | ## |
| 30 | | ## Grammar definition file (DFA and dictionary) |
| 31 | | ## |
| 32 | | |
| 33 | | #### There are three ways to specify the grammar files. |
| 34 | | #### (1) and (2) can be used multiple times. |
| 35 | | |
| 36 | | #### (1) Specify by common prefix of .dfa and .dict files. Comma-separated |
| 37 | | #### prefixes can be specified for multiple grammar recognition |
| 38 | | #-gram /cdrom/testrun/sample_grammars/vfr/vfr |
| 39 | | |
| 40 | | #### (2) Or you can give Julian a text file which contains list of grammar |
| 41 | | #### prefixes one per line. |
| 42 | | #-gramlist file |
| 43 | | |
| 44 | | #### (3) Classic way to specify a grammar. |
| 45 | | -dfa sample.dfa |
| 46 | | -v sample.dict |
| 47 | | |
| 48 | | #### If you want to clear previously specified grammars, use this at the |
| 49 | | #### point. |
| 50 | | #-nogram |
| 51 | | |
| 52 | | ## |
| 53 | | ## Acoustic HMM file |
| 54 | | ## |
| 55 | | # support ascii hmmdefs or binary format (converted by "mkbinhmm") |
| 56 | | # format (ascii/binary) will be automatically detected |
| 57 | | -h hmmdefs |
| 58 | | |
| 59 | | ## triphone model needs HMMList that maps logical triphone to physical ones. |
| 60 | | -hlist tiedlist |
| 61 | | |
| 62 | | ###################################################################### |
| 63 | | #### Multiple grammar recognition |
| 64 | | ###################################################################### |
| 65 | | #-multigramout # Output results for each grammar |
| 66 | | |
| 67 | | ###################################################################### |
| 68 | | #### Language Model |
| 69 | | ###################################################################### |
| 70 | | ## |
| 71 | | ## word insertion penalty |
| 72 | | ## |
| 73 | | #-penalty1 0.0 # first pass |
| 74 | | #-penalty2 0.0 # second pass |
| 75 | | |
| 76 | | ###################################################################### |
| 77 | | #### Dictionary |
| 78 | | ###################################################################### |
| 79 | | ## |
| 80 | | ## do not giveup startup on error words |
| 81 | | ## |
| 82 | | #-forcedict |
| 83 | | |
| 84 | | ###################################################################### |
| 85 | | #### Acoustic Model |
| 86 | | ###################################################################### |
| 87 | | ## |
| 88 | | ## Context-dependency handling will be enabled according to the model type. |
| 89 | | ## Try below if julius wrongly detect the type of hmmdefs |
| 90 | | ## |
| 91 | | #-no_ccd # disable context-dependency handling |
| 92 | | #-force_ccd # enable context-dependency handling |
| 93 | | |
| 94 | | ## |
| 95 | | ## If julius go wrong with checking parameter type, try below. |
| 96 | | ## |
| 97 | | #-notypecheck |
| 98 | | # |
| 99 | | |
| 100 | | ## |
| 101 | | ## (PTM/triphone) switch computation method of IWCD on 1st pass |
| 102 | | ## |
| 103 | | #-iwcd1 best N # assign average of N-best likelihood of the same context |
| 104 | | -iwcd1 max # assign maximum likelihood of the same context |
| 105 | | #-iwcd1 avg # assign average likelihood of the same context (default) |
| 106 | | |
| 107 | | ###################################################################### |
| 108 | | #### Gaussian Pruning |
| 109 | | ###################################################################### |
| 110 | | ## Number of mixtures to select in a mixture pdf. |
| 111 | | ## This default value is optimized for IPA99's PTM, |
| 112 | | ## with 64 Gaussians per codebook |
| 113 | | #-tmix 2 |
| 114 | | |
| 115 | | ## Select Gaussian pruning algorithm |
| 116 | | ## defulat: beam (standard setting), safe (others) |
| 117 | | -gprune safe # safe pruning, accurate but slow |
| 118 | | #-gprune heuristic # heuristic pruning |
| 119 | | #-gprune beam # beam pruning, fast but sensitive |
| 120 | | #-gprune none # no pruning |
| 121 | | |
| 122 | | ###################################################################### |
| 123 | | #### Gaussian Mixture Selection |
| 124 | | ###################################################################### |
| 125 | | #-gshmm hmmdefs # monophone HMM for GMS |
| 126 | | # (OFF when not specified) |
| 127 | | #-gsnum 24 # number of states to be selected on GMS |
| 128 | | |
| 129 | | ###################################################################### |
| 130 | | #### Search Parameters |
| 131 | | ###################################################################### |
| 132 | | #-b 2000 # beam width on 1st pass (#nodes) for monophone |
| 133 | | #-b 800 # beam width on 1st pass (#nodes) for triphone,PTM |
| 134 | | #-b 1000 # beam width on 1st pass (#nodes) for triphone,PTM,engine=v2.1 |
| 135 | | #-b2 100 # beam width on 2nd pass (#words) |
| 136 | | #-sb 80.0 # score beam envelope threshold |
| 137 | | #-s 500 # hypotheses stack size on 2nd pass (#hypo) |
| 138 | | #-m 2000 # hypotheses overflow threshold (#hypo) |
| 139 | | #-lookuprange 5 # lookup range for word expansion (#frame) |
| 140 | | #-n 1 # num of sentences to find (#sentence) |
| 141 | | #-n 10 # (default for 'standard' configuration) |
| 142 | | #-output 1 # num of found sentences to output (#sentence) |
| 143 | | #-looktrellis # search within only backtrellis words |
| 144 | | |
| 145 | | ###################################################################### |
| 146 | | #### Inter-word Short Pause Handling |
| 147 | | ###################################################################### |
| 148 | | ## |
| 149 | | ## Specify short pause model name to be treated as special |
| 150 | | ## |
| 151 | | #-spmodel "sp" # HMM model name |
| 152 | | |
| 153 | | ## |
| 154 | | ## For insertion of context-free short-term inter-word pauses between words |
| 155 | | ## (multi-path version only) |
| 156 | | ## |
| 157 | | #-iwsp # append a skippable sp model at all word ends |
| 158 | | #-iwsppenalty 0.0 # transition penalty for the appenede sp models |
| 159 | | |
| 160 | | ###################################################################### |
| 161 | | #### Speech Input Source |
| 162 | | ###################################################################### |
| 163 | | ## select one (default: mfcfile) |
| 164 | | #-input mfcfile # MFCC file in HTK parameter file format |
| 165 | | #-input rawfile # raw wavefile (auto-detect format) |
| 166 | | # WAV(16bit) or |
| 167 | | # RAW(16bit(signed short),mono,big-endian) |
| 168 | | # AIFF,AU (with libsndfile extension) |
| 169 | | # other than 16kHz, sampling rate should be specified |
| 170 | | # by "-smpFreq" option |
| 171 | | #-input mic # direct microphone input |
| 172 | | # device name can be specified via env. val. "AUDIODEV" |
| 173 | | #-input netaudio -NA host:0 # direct input from DatLink(NetAudio) host |
| 174 | | #-input adinnet -adport portnum # via adinnet network client |
| 175 | | #-input stdin # from standard tty input (pipe) |
| 176 | | |
| 177 | | #-filelist filename # specify file list to be recognized in batch mode |
| 178 | | |
| 179 | | #-nostrip # switch OFF dropping of invalid input segment. |
| 180 | | # (default: strip off invalid segment (0 sequence etc.) |
| 181 | | #-zmean # enable DC offset removal (invalid for mfcfile input) |
| 182 | | |
| 183 | | ###################################################################### |
| 184 | | #### Recording |
| 185 | | ###################################################################### |
| 186 | | #-record directory # auto-save recognized speech data into the dir |
| 187 | | |
| 188 | | ###################################################################### |
| 189 | | #### GMM-based Input Verification and Rejection |
| 190 | | ###################################################################### |
| 191 | | #-gmm gmmdefs # specify GMM definition file in HTK format |
| 192 | | #-gmmnum 10 # num of Gaussians to be computed per mixture |
| 193 | | #-gmmreject "noise,laugh,cough" # list of GMM names to be rejected |
| 194 | | |
| 195 | | ###################################################################### |
| 196 | | #### Too Short Input Rejection |
| 197 | | ###################################################################### |
| 198 | | -rejectshort 2000 # reject input shorter than specified millisecond |
| 199 | | |
| 200 | | ###################################################################### |
| 201 | | #### Speech Detection |
| 202 | | ###################################################################### |
| 203 | | #-pausesegment # turn on speech detection by level and zero-cross |
| 204 | | #-nopausesegment # turn off speech detection by level and zero-cross |
| 205 | | # (default: on for mic or adinnet, off for file) |
| 206 | | #-lv 2000 # threshold of input level (0-32767) |
| 207 | | #-headmargin 300 # head margin of input segment (msec) |
| 208 | | #-tailmargin 400 # tail margin of input segment (msec) |
| 209 | | #-zc 60 # threshold of number of zero-cross in a second |
| 210 | | |
| 211 | | ###################################################################### |
| 212 | | #### Acoustic Analysis |
| 213 | | ###################################################################### |
| 214 | | -smpFreq 8000 # sampling rate (Hz) |
| 215 | | #-smpPeriod 625 # sampling period (ns) (= 10000000 / smpFreq) |
| 216 | | #-fsize 400 # window size (samples) |
| 217 | | #-fshift 160 # frame shift (samples) |
| 218 | | #-delwin 2 # delta window (frames) |
| 219 | | #-hifreq -1 # cut-off hi frequency (Hz) (-1: disable) |
| 220 | | #-lofreq -1 # cut-off low frequency (Hz) (-1: disable) |
| 221 | | #-cmnsave filename # save CMN param to file (update per input) |
| 222 | | #-cmnload filename # load initial CMN param from file on startup |
| 223 | | |
| 224 | | ###################################################################### |
| 225 | | #### Spectral Subtraction (SS) |
| 226 | | ###################################################################### |
| 227 | | #-sscalc # do SS using head silence (file input only) |
| 228 | | #-sscalclen 300 # length of head silence for SS (msec) |
| 229 | | #-ssload filename # load constant noise spectrum from file for SS |
| 230 | | #-ssalpha 2.0 # alpha coef. for SS |
| 231 | | #-ssfloor 0.5 # spectral floor for SS |
| 232 | | |
| 233 | | ###################################################################### |
| 234 | | #### Forced alignment |
| 235 | | ###################################################################### |
| 236 | | #-walign # do forced alignment with result per word |
| 237 | | #-palign # do forced alignment with result per phoneme |
| 238 | | #-salign # do forced alignment with result per HMM state |
| 239 | | |
| 240 | | ###################################################################### |
| 241 | | #### Word Confidence Scoring |
| 242 | | ###################################################################### |
| 243 | | #-cmalpha 0.05 # smoothing coef. alpha |
| 244 | | |
| 245 | | ###################################################################### |
| 246 | | #### Output |
| 247 | | ###################################################################### |
| 248 | | #-separatescore # output language and acoustic score separately |
| 249 | | #-progout # output partial result per a time interval |
| 250 | | #-proginterval 300 # time interval for "-progout" (msec) |
| 251 | | #-quiet # output minimal result |
| 252 | | #-demo # = "-progout -quiet", suitable for dictation demo |
| 253 | | #-debug # output full message for debug |
| 254 | | #-charconv from to # output character set conversion (see manual for |
| 255 | | # available code set name) |
| 256 | | |
| 257 | | ###################################################################### |
| 258 | | #### Server module mode |
| 259 | | ###################################################################### |
| 260 | | #-module # Run Julius on "Server module mode" |
| 261 | | #-module 5530 # (when using another port number for connection) |
| 262 | | #-outcode WLPSC # select output message toward module (WLPSCwlps) |
| 263 | | |
| 264 | | ###################################################################### |
| 265 | | #### Misc. |
| 266 | | ###################################################################### |
| 267 | | #-help # output help and exit |
| 268 | | #-setting # output engine configuration and exit |
| 269 | | #-C jconffile # expand other jconf file in its place |
| 270 | | |
| 271 | | ################################################################# end of file |
| 272 | | |
| 273 | | }}} |
