4.2.1 Network Decoding:
Command Line Options and Arguments
Virtually all of our significant speech recognition functionality
is encapsulated in a single program, called isip_recognize.
This program has many options, arguments, and modes. In this section,
we discuss those options that are relevant to network decoding.
For a general discussion of our command line
interface, see our foundation class documentation on
utilities.
This will teach you the basics about how command line parameters
are parsed, and explain some general options that are available
to all utilities (e.g., -debug).
To get started, execute the command "isip_recognize -help".
You will obtain a list of the most important available options,
including the ones discussed in this section.
isip221_[1]: isip_recognize -help
name: isip_recognize
synopsis: isip_recognize -param param.sof [options] [file1.sof...]
description: a speech recognizer that transforms speech to text
example: isip_recognize -param param.sof -list identifier_list.sof
Options:
-directory: directory to place output (default = source file directory)
-extension: new file extension (default = sof)
-list: an input list of utterance identifiers
-output_file: contains recognition results (used in file output mode)
-output_format: output file format (e.g., database or NIST format)
-output_mode: controls how the data is output (e.g., to a file or list)
-parameter_file: specifies the parameter file that configures the recognizer
-preserve: number of levels of subdirectories to preserve (default = 0)
-suffix: new basename suffix (default = none)
-type: text or binary (default = binary)
-verify: run without actual decoding
-verbosity: verbosity level (default = NONE)
arguments: sdb or sof files can be passed as arguments
man page: $ISIP/util/speech/isip_transform/index.html
For a more detailed explanation of the isip_recognize utility, visit
its manual page.
Many options are common to all of our tools. Options such as -debug,
-help, and -version have obvious purposes and are used for every tool.
The help messages usually omit these options since they are so
common. Other options such as -directory, -extension, -preserve, etc.
are explained in
Section 3.3.2
for isip_transform. Clicking on one of these options above will link
you to the option's desciption in that section. The
CommandLine
class is responsible for reserving these option names. For a more complete
description of the CommandLine class,
click here. The purpose of this section is to explain and
demonstrate each of the options that are unique to isip_recognize. The
option's explanation will be followed by an example in order to
further demonstrate the purpose of each option.
Option: -output_file
The desired output filename can be specified. This example directs the
output into a file named out.tmp
instead of
hypo.out which is
set in our parameter file,
params_decode.sof.
isip_recognize -verbose ALL -parameter_file params_decode.sof
-output_file out.tmp -list identifiers.sof
The screen output will be:
isip106_[1]: isip_recognize -verbose ALL -parameter_file params_decode.sof -output_file out.tmp -list identifiers.sof
Command: isip_recognize -verbose ALL -parameter_file params_decode.sof -output_file out.tmp -list identifiers.sof
Version: 1.23 (not released) 2003/05/21 23:10:45
loading audio database: $TUTORIAL/exp/exp_00/research/isip/audio_db.sof
*** no transcription database file was specified ***
loading front-end: $TUTORIAL/config/frontend.sof
loading language model: $TUTORIAL/models/lm_tidigits.sof
loading statistical model pool: $TUTORIAL/models/smp_tidigits.sof
loading configuration file: $TUTORIAL/config/tidigits_config.sof
opening the output file: out.tmp
processing file 1 (ah_954a): $TUTORIAL/features/ah_954a.sof
hyp: SILENCE NINE FIVE FOUR SILENCE
score: -9310.9453125 frames: 146
processing file 2 (ah_o321oo7a): $TUTORIAL/features/ah_o321oo7a.sof
hyp: SILENCE OH THREE TWO PAUSE ONE OH OH SEVEN SILENCE
score: -18112.123046875 frames: 287
processed 2 file(s) successfully, attempted 2 file(s), 433 frame(s)
Note that if the output filename is not set, either in the parameter file
or on by -output_file, the output WILL NOT be saved in a file.
Option: -output_format
This tells isip_recognize to generate output as either a DATABASE,
or a NIST_SCORE format. For instance:
isip_recognize -output_file out.tmp -output_format DATABASE -parameter_file params_decode.sof -list identifiers.sof
isip106_[1]: isip_recognize -verbose ALL -output_file out_datab.tmp -output_format DATANASE -parameter_file params_decode.sof -list identifiers.sof
Command: isip_recognize -verbose ALL -output_file out_datab.tmp -output_format DATABASE -parameter_file params_decode.sof -list identifiers.sof
Version: 1.23 (not released) 2003/05/21 23:10:45
loading audio database: $TUTORIAL/exp/exp_00/research/isip/audio_db.sof
*** no transcription database file was specified ***
loading front-end: $TUTORIAL/config/frontend.sof
loading language model: $TUTORIAL/models/lm_tidigits.sof
loading statistical model pool: $TUTORIAL/models/smp_tidigits.sof
loading configuration file: $TUTORIAL/config/tidigits_config.sof
opening the output file: out_datab.tmp
processing file 1 (ah_954a): $TUTORIAL/features/ah_954a.sof
hyp: SILENCE NINE FIVE FOUR SILENCE
score: -9310.9453125 frames: 146
processing file 2 (ah_o321oo7a): $TUTORIAL/features/ah_o321oo7a.sof
hyp: SILENCE OH THREE TWO PAUSE ONE OH OH SEVEN SILENCE
score: -18112.123046875 frames: 287
processed 2 file(s) successfully, attempted 2 file(s), 433 frame(s)
The command will generate the file
out_datab.tmp. Open
this file and confirm that its format is Sof database.
This is isip_recognize's default format. The annotation graph
format has not yet been implemented. It will be invoked by:
isip_recognize -output_format SOF
-parameter_file params_decode.sof -list identifiers.sof
Option: -output_mode
The output_mode option sets the output mode specification and can be
FILE, LIST, or TRANSFORM.
- FILE: outputs hypotheses to a single output file, and is the default
option
isip_recognize -output_mode FILE -output_file out.tmp
-parameter_file params_decode.sof -list identifiers.sof
This sets the output mode to FILE and puts the output into
out.tmp.
The screen output will be:
isip106_[1]: isip_recognize -verbose ALL -output_file out_file.tmp -parameter_file params_decode.sof -output_mode FILE -list identifiers.sof
Command: isip_recognize -verbose ALL -output_file out_file.tmp -parameter_file params_decode.sof -output_mode FILE -list identifiers.sof
Version: 1.23 (not released) 2003/05/21 23:10:45
loading audio database: $TUTORIAL/exp/exp_00/research/isip/audio_db.sof
*** no transcription database file was specified ***
loading front-end: $TUTORIAL/config/frontend.sof
loading language model: $TUTORIAL/models/lm_tidigits.sof
loading statistical model pool: $TUTORIAL/models/smp_tidigits.sof
loading configuration file: $TUTORIAL/config/tidigits_config.sof
opening the output file: out_file.tmp
processing file 1 (ah_954a): $TUTORIAL/features/ah_954a.sof
hyp: SILENCE NINE FIVE FOUR SILENCE
score: -9310.9453125 frames: 146
processing file 2 (ah_o321oo7a): $TUTORIAL/features/ah_o321oo7a.sof
hyp: SILENCE OH THREE TWO PAUSE ONE OH OH SEVEN SILENCE
score: -18112.123046875 frames: 287
processed 2 file(s) successfully, attempted 2 file(s), 433 frame(s)
The FILE mode is the default output mode for isip_recognize.
- LIST: outputs each hypothesis to a different file in a list specified
using the option -output_list. In the example below, the
file
list.sof
contains a list of two files. The hypotheses for both utterances will
be placed in a seperate file. Run the command below:
isip_recognize -verbose ALL -parameter_file params_decode.sof -output_mode LIST -output_list list.sof -list identifiers.sof
This command will generate two files:
out_list1.tmp and
out_list2.tmp. Make
sure that the number of output files in your list matches the number
of utterances you are decoding.
- TRANSFORM: outputs each hypothesis as a Sof file and puts
the output in a directory specified by
-directory
If -directory
is not set, the output will, by default, be placed in the same directory
as the input features. The output filename will match the feature file
name and inckude a "_0" suffix. In other words, if one utterance to be
decoded is "data/bg_1647a.sof" its output will be "data/bg_1647a_0.sof".
This example will define -directory. Run the command
isip_recognize -verbose ALL -parameter_file params_decode.sof -output_mode TRANSFORM -directory ./ -list identifiers.sof
This command generates the two files
ah_954a.sof,
ah_o321oo7a.sof.
Option: -verify
The -verify option allows the checks the configuration of isip_recognize,
but does not actually process each utterance. When -verify is used,
isip_recognize checks that each specified configuration is in
place and that all of the data files exist.
This example verifies that all parameters are correct and that files exist:
Option: -verbosity
This option controls the amount of screen output that is generated.
Controlling the output is helpful when configuring and tweaking
parameters. The examples below show how the different -verbosity
levels affect the screen output.
NONE:
isip_recognize -verbose NONE -parameter_file params_decode.sof
-output_file out.tmp -list identifiers.sof
This command does not generate any output. Also, if a verbosity level
is not specified, no screen output will generated.
BRIEF:
isip_recognize -verbose BRIEF -parameter_file params_decode.sof -output_file out.tmp -list identifiers.sof
Output:
isip106_[1]: isip_recognize -verbose BRIEF -output_file out_file.tmp -parameter_file params_decode.sof -list identifiers.sof
Command: isip_recognize -verbose BRIEF -output_file out.tmp -parameter_file params_decode.sof -list identifiers.sof
Version: 1.23 (not released) 2003/05/21 23:10:45
loading audio database: $TUTORIAL/exp/exp_00/research/isip/audio_db.sof
*** no transcription database file was specified ***
loading front-end: $TUTORIAL/config/frontend.sof
loading language model: $TUTORIAL/models/lm_tidigits.sof
loading statistical model pool: $TUTORIAL/models/smp_tidigits.sof
loading configuration file: $TUTORIAL/config/tidigits_config.sof
opening the output file: out_file.tmp
processing file 1 (ah_954a): $TUTORIAL/features/ah_954a.sof
hyp: SILENCE NINE FIVE FOUR SILENCE
score: -9310.9453125 frames: 146
processing file 2 (ah_o321oo7a): $TUTORIAL/features/ah_o321oo7a.sof
hyp: SILENCE OH THREE TWO PAUSE ONE OH OH SEVEN SILENCE
score: -18112.123046875 frames: 287
processed 2 file(s) successfully, attempted 2 file(s), 433 frame(s)
Clearly, more output is generated. If a problem were to occur during
the recognition process, the source of the problem would be easier to
find.
ALL:
isip_recognize -verbose ALL -parameter_file params_decode.sof
-output_file out.tmp -list identifiers.sof
ALL produces the same amount of output as BRIEF.
|