root/Trunk/Scripts/AcousticModel_scripts/HTK/manual_testing/julius.jconf

# !!!!!! notes: need a 2 gram model in addition to the 3 gram model to ge this to work!!!
# and both models need  to be trained on the same data!
#
# Copyright (c) 1991-2006 Kawahara Lab., Kyoto University
# Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and Technology
# Copyright (c) 2005-2006 Julius project team, Nagoya Institute of Technology
# All rights reserved
#
# Example of Runtime Configuration File for Julius rev.3.5
#
# 1) NOTICE: relative paths must be relative to THIS FILE.
#            Current working directory does not affect.
# 2) Blank lines and comments after '#' are ignored.
#    Use `\#' to specify `#' in an argument, 
# 3) Each line should be shorter than 512 bytes.
# 4) Syntax of each arguments are same as corresponding command options.
# 5) The commented-out values in this file are default values in Julius.
#

######################################################################
#### Files
######################################################################
##
## Language model file(s)
##
# in ARPA standard format
#-nlr /cdrom/lang_m/20k/MNP.111.20k.1.arpa.gz           # LR 2-gram
#-nrl /cdrom/lang_m/20k/MNP.111.20k.rev.1-1.10p.arpa.gz # RL 3-gram
# or in binary format (generated by "mkbingram"), 
#-d /cdrom/lang_m/20k/bingram_for_julius/MNP.111.20k.1-1.10p.bingram.gz
-nlr ../../../LanguageModels/sphinx_models_hub4opensrc_jan2002/bn.bigram.arpa
-nrl  ../../../LanguageModels/sphinx_models_hub4opensrc_jan2002/bn99_64000_lm.arpa

##
## Word dictionary file
##
#-v /cdrom/lang_m/20k/MNP.111.20k.htkdic
-v ../../../LanguageModels/sphinx_models_hub4opensrc_jan2002/LM.vocabulary_julius


##
## Acoustic HMM file
##
# support ascii hmmdefs or binary format (converted by "mkbinhmm")
# format (ascii/binary) will be automatically detected
#-h /cdrom/phone_m/jnas/ptm-3000x64/gid/hmmdefs,tmix.gz # HTK format
-h ../../../AcousticModels/HTK/16kHz:16-bit/MFCC_O_D/hmmdefs

## triphone model needs HMMList that maps logical triphone to physical ones.
#-hlist /cdrom/phone_m/jnas/logicalTri
-hlist ../../../AcousticModels/HTK/16kHz:16-bit/MFCC_O_D/tiedlist


######################################################################
#### Language Model 
######################################################################
##
## Language score weights and insertion penalty
##
## for example, "-lmp 8.0 7.0" indicates that the language log probability
## 'log p(w)' should be translated as '(log p(w)) * 8.0 + 7.0' in search.
##
##default values
##for monophone
#-lmp 5.0 -1.0          # 1st pass(2-gram)
#-lmp2 6.0 0.0          # 2nd pass(3-gram)
##for triphone
#-lmp 8.0 -2.0
#-lmp2 8.0 -2.0
##for triphone with setup='v2.1', when IWCD not handled on 1st pass
#-lmp 9.0 8.0
#-lmp2 11.0 -2.0

##
## additional penalty for transparent words
##
#-transp 0.0

######################################################################
#### Dictionary
######################################################################
##
## name of beginning-of-sentence word in dictionary
## 
#-silhead '<s>'

##
## name of end-of-sentence word in dictionary
##
#-siltail '</s>'

##
## do not giveup startup on error words
##
#-forcedict

######################################################################
#### Acoustic Model
######################################################################
##
## 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

## 
## If julius go wrong with checking parameter type, try below.
##
#-notypecheck
#

##
## (PTM/triphone) switch computation method of IWCD on 1st pass
##
#-iwcd1 best 3  # assign average of N-best likelihood of the same context (default, N=3)
#-iwcd1 max     # assign maximum likelihood of the same context
#-iwcd1 avg     # assign average likelihood of the same context

######################################################################
#### 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
#-tmix 2

## Select Gaussian pruning algorithm
## defulat: beam (standard setting), safe (others)
#-gprune safe           # safe pruning, accurate but slow
#-gprune heuristic      # heuristic pruning
#-gprune beam           # beam pruning, fast but sensitive
#-gprune none           # no pruning

######################################################################
#### Gaussian Mixture Selection 
######################################################################
#-gshmm hmmdefs         # monophone HMM for GMS
                        # (OFF when not specified)
#-gsnum 24              # number of states to be selected on GMS

######################################################################
#### Search Parameters
######################################################################
#-b 400                 # beam width on 1st pass (#nodes) for monophone
#-b 800                 # beam width on 1st pass (#nodes) for triphone,PTM
#-b 1000                # beam width on 1st pass (#nodes) for triphone,PTM,engine=v2.1
#-b2 30                 # beam width on 2nd pass (#words)
#-sb 80.0               # score beam envelope threshold
#-s 500                 # hypotheses stack size on 2nd pass (#hypo)
#-m 2000                # hypotheses overflow threshold (#hypo)
#-lookuprange 5         # lookup range for word expansion (#frame)
#-n 1                   # num of sentences to find (#sentence)
#-n 10                  #   (default for 'standard' configuration)
#-output 1              # num of found sentences to output (#sentence)

### when configured with "./configure --enable-lowmem"
#-iwcache 10            # % of inter-word LM cache size
                        # 100 is equivalent with no "--enable-lowmem"

### when configured with "--enable-lowmem2"
#-sepnum 150            # num of high freq words to be separated from tree

######################################################################
#### Graph Output (--enable-graphout)
######################################################################
##
## Merge same words in graph.
##  -1: not merge, leave same words on the same location with diff. score
##   0: merge same words at same location
##  >0: merge same words around the margin
## 
#-graphrange 0

##
## Graph depth cutting threshold (num of words per reference word).
## (-1: disable)
##
#-graphcut 80

##
## Maximum number of boundary adjustment loop.
##
#-graphboundloop 20

##
## When "-graphsearchdelay" option is set, Julius modifies its alogrithm of
## graph generation on the 2nd pass not to apply search termination by graph
## merging until the first sentence candidate is found.
##
## This option may result in slight improvement of graph accuracy only
## when you are going to generate a huge word graph by setting broad search.
## Namely, it may result in better graph accuracy when you set wide beams on
## both 1st pass "-b" and 2nd pass "-b2", and large number for "-n".
##
#-graphsearchdelay
#-nographsearchdelay  # this is default (off)

######################################################################
#### Inter-word Short Pause Handling
######################################################################
##
## Specify short pause model name to be treated as special
##
#-spmodel "sp"          # HMM model name

##
## Add a pause word entry to the dictionary
##
#-iwspword
#-iwspentry "<UNK> [sp] sp sp"  # default word entry to be added

##
## 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
#-iwsppenalty 0.0       # transition penalty for the appenede sp models

######################################################################
#### Short-pause Segmentation  (--enable-sp-segment)
######################################################################
#-spdur 10              # sp duration frame on 1st pass
                        # Input will be segmented if sp word becomes top
                        # during this period

######################################################################
#### Speech Input Source
######################################################################
## 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 filename     # specify file list to be recognized in batch mode

#-nostrip               # switch OFF dropping of invalid input segment.
                        # (default: strip off invalid segment (0 sequence etc.)
#-zmean                 # enable DC offset removal (invalid for mfcfile input)
#-zmeanframe            # frame-wise DC offset removal (same as HTK)

######################################################################
#### Recording
######################################################################
#-record directory      # auto-save recognized speech data into the dir

######################################################################
#### GMM-based Input Verification and Rejection
######################################################################
#-gmm gmmdefs           # specify GMM definition file in HTK format
#-gmmnum 10             # num of Gaussians to be computed per mixture
#-gmmreject "noise,laugh,cough" # list of GMM names to be rejected

######################################################################
#### Too Short Input Rejection
######################################################################
#-rejectshort 800       # reject input shorter than specified millisecond

######################################################################
#### Speech Detection
######################################################################
#-pausesegment          # turn on speech detection by level and zero-cross
#-nopausesegment        # turn off speech detection by level and zero-cross
                        # (default: on for mic or adinnet, off for file)
#-lv 2000               # threshold of input level (0-32767)
#-headmargin 300        # head margin of input segment (msec)
#-tailmargin 400        # tail margin of input segment (msec)
#-zc 60                 # threshold of number of zero-cross in a second

######################################################################
#### Acoustic Analysis
######################################################################
#-smpFreq 16000         # sampling rate (Hz)
#-smpPeriod 625         # sampling period (ns) (= 10000000 / smpFreq)
#-fsize 400             # window size (samples)
#-fshift 160            # frame shift (samples)
#-preemph 0.97          # pre-emphasis coef.
#-fbank 24              # number of filterbank channels
#-ceplif 22             # cepstral liftering coef.
#-rawe                  # use raw energy
#-norawe                # not use raw energy (this is default)
#-enormal               # normalize log energy
#-noenormal             # not normalize log energy (this is default)
#-escale 1.0            # scaling log energy for enormal
#-silfloor 50.0         # energy silence floor in dB for enormal
#-delwin 2              # delta window (frames)
#-accwin 2              # acceleration window (frames)
#-hifreq -1             # cut-off hi frequency (Hz) (-1: disable)
#-lofreq -1             # cut-off low frequency (Hz) (-1: disable)

######################################################################
#### Cepstral Mean Normalization (CMN)
######################################################################
#-cmnsave filename      # save CMN param to file (update per input)
#-cmnload filename      # load initial CMN param from file on startup
#-cmnmapweight 100.0    # weight for MAP-CMN
#-cmnnoupdate           # keep initial CMN, not updating from input
                        # (use with -cmnload)

######################################################################
#### Spectral Subtraction (SS)
######################################################################
#-sscalc                # do SS using head silence (file input only)
#-sscalclen 300         # length of head silence for SS (msec)
#-ssload filename       # load constant noise spectrum from file for SS
#-ssalpha 2.0           # alpha coef. for SS
#-ssfloor 0.5           # spectral floor for SS

######################################################################
#### Forced alignment
######################################################################
#-walign                # do forced alignment with result per word 
#-palign                # do forced alignment with result per phoneme
#-salign                # do forced alignment with result per HMM state

######################################################################
#### Word Confidence Scoring
######################################################################
#-cmalpha 0.05          # smoothing coef. alpha

######################################################################
#### Output 
######################################################################
#-separatescore         # output language and acoustic score separately
#-progout               # output partial result per a time interval
#-proginterval 300      # time interval for "-progout" (msec)
#-quiet                 # output minimal result
#-demo                  # = "-progout -quiet", suitable for dictation demo
#-debug                 # output full message for debug
#-charconv from to      # output character set conversion (see manual for
                        # available code set name)

######################################################################
#### Server module mode
######################################################################
#-module                # Run Julius on "Server module mode"
#-module 5530           # (when using another port number for connection)
#-outcode WLPSC         # select output message toward module (WLPSCwlps)

######################################################################
#### Misc.
######################################################################
#-help                  # output help and exit
#-setting               # output engine configuration and exit
#-C jconffile           # expand other jconf file in its place

################################################################# end of file