These notes describe the process of testing a set of neural nets using the ASAT TestBed software. The ASAT TestBed software is currently designed to use the TIMIT data set and neural network classifiers built on QuickNet software.

Testing

test.pl is the training script. For this step, you must have a set of weight files trained by the train.pl script and feature files and label files as built by the TestBed software. Currently, the test.pl script does not implement a batch flag as the rest of the TestBed software does, so each different weight file must be tested by a separate run of the TestBed software. The sample shell scripts 

run_test.feat#.sh
show how this works.

The configuration files for the test.pl software work almost identically to the configuration files for the train.pl script. The required keys are: 

[global]
# In addition to the required keys from the Build programs, the following
# new key is required
outputdir = [ output directory for the trained weight files ]

[test]
normfile = [ normalization file for the weight file to be used (QuickNet)]
testfeaslist = [ list of feature files that may be used in testing ]
testlabslist = [ list of associated feature labels for testing validation ]
weightfile = [ weight file to use for testing ]


[tstprg]
# These are parameters to pass to the underlying training program.  For 
# QuickNet neural nets, the following parameters are typical.  Output size
# and hidden size should vary by feature.
hardtarget_format=ilab
hardtarget_window_offset=4
window_extent=9
ftr1_window_len=9
ftr1_ftr_count=26
ftr1_norm_mode=file
mlp3_input_size=234
mlp3_output_size=5
mlp3_hidden_size=100
mlp3_bunch_size=16