Index of /sphinx/models/lm/turtle-class-lm

Name	Last modified	Size

Parent Directory		-
README.html	06-Nov-2009 23:08	3.9K
domain.corpus	06-Nov-2009 23:08	67
domain.ctl	06-Nov-2009 23:08	67
domain.def	06-Nov-2009 23:08	345
domain.dic	06-Nov-2009 23:08	924
domain.lm	06-Nov-2009 23:08	868
domain.sent	06-Nov-2009 23:08	84
domain.sent.arpabo	06-Nov-2009 23:08	868
domain.token	06-Nov-2009 23:08	48
domain.vocab	06-Nov-2009 23:08	38
domain.words	06-Nov-2009 23:08	261
sphinx2-simple-class	06-Nov-2009 23:08	726
turtle-class-lm-2001-03-29.tgz	06-Nov-2009 23:08	3.0K

Turtle Class LM Example

Sphinx2 Turtle Class Language Model Example

Kevin A. Lenzo (lenzo@cs.cmu.edu)
March 2, 2001
v0.01

This is an example 'turtle' domain with a class-based language model. For this example, the contents domain.corpus were given to the lmtool web-based language model building tool:


GO [direction] [number] [unit]

ROTATE [direction] [number] [unit]

This will cover a small language of things like GO FORWARD TEN METERS and ROTATE RIGHT FORTY FIVE DEGREES.

After the corpus is run through the lmtool, the tarball is unpacked. Here, the relevant files have all been renamed to domain.something, where the something is a typical three-letter extension. sphinx2-simple-class is a shell script to invoke the sphinx2-continuous executable with an appropriate argument list:


#!/bin/sh

S2CONTINUOUS=/usr/local/bin/sphinx2-continuous

HMM=/usr/local/share/sphinx2/model/hmm/6k

TASK=.

$S2CONTINUOUS -live TRUE -ctloffset 0 -ctlcount 100000000 -cepdir ${TASK}/ctl -datadir ${TASK}/ctl -agcmax FALSE -langwt 6.5 -fwdflatlw 8.5 -rescorelw 9.5 -ugwt 0.5 -fillpen 1e-10 -silpen 0.005 -inspen 0.65 -top 1 -topsenfrm 3 -topsenthresh -70000 -beam 2e-06 -npbeam 2e-06 -lpbeam 2e-05 -lponlybeam 0.0005 -nwbeam 0.0005 -fwdflat FALSE -fwdflatbeam 1e-08 -fwdflatnwbeam 0.0003 -bestpath TRUE -kbdumpdir ${TASK} -lmctlfn ${TASK}/domain.ctl -dictfn ${TASK}/domain.dic -ndictfn ${HMM}/noisedict -phnfn ${HMM}/phone -mapfn ${HMM}/map -hmmdir ${HMM} -hmmdirlist ${HMM} -8bsen TRUE -sendumpfn ${HMM}/sendump -cbdir ${HMM}

In particular, note the following arguments:

-lmctlfn domain.ctl
Language model control file name. See domain.ctl as an example.
-dictfn domain.dic
Dictionary file name; here, domain.dic.
The control file, domain.ctl looks like this:

{ domain.def } domain.lm general { [direction] [number] [unit] }
The class definition file is the first item in braces (domain.def) class definition file. This can be a fully qualified path, such as
```
    { /usr0/lenzo/task/domain.def }
   
```
The path to the language model file follows on the next line. The example here assumes that domain.lm is an ARPA back-off language model (arpabo model) in the current working directory.
The symbolic name of the language model is general.
There are three classes declared, that are described in the .def file:
[direction], [number], and [unit].

 The domain.def file might 
  look like the following (note that there is no white space within
the tokens):

  
LMCLASS [direction]
FORWARD
BACKWARD
RIGHT
LEFT
END [direction]

LMCLASS [number]
ONE
TWO
THREE
FOUR
FIVE
SIX
SEVEN
EIGHT
NINE
TEN
ELEVEN
TWELVE
THIRTEEN
FOURTEEN
FIFTEEN
SIXTEEN
SEVENTEEN
EIGHTEEN
NINETEEN
TWENTY
THIRTY
FORTY
FIFTY
SIXY
SEVENTY
EIGHTY
NINETY
HUNDRED
THOUSAND
END [number]

LMCLASS [unit]
METER
METERS
DEGREE
DEGREES
END [unit]