This experiment was for replicating the table 17 in the following document.

http://www.cavs.msstate.edu/hse/ies/publications/reports/aurora_frontend/2002/report_012202_v21.pdf

The difference between the setup for the experiment in the document and the experiment I ran was the number of mixtures. The results I got were much better. The main reason was that I used 8 mixtures. A difference of almost 155 was observed for some sets. The data for training and testing can be obtained from CD's in a folder named AURORA.

The feature files, lists and the results are here:

/cavs/hse/ies/isip/data/research/nonlinear/exp_011/aurora/asr_wsj_tutorial_v1.4.2/exp_00

The decode_* has the results for different sets, where * denotes the set number.
The results obtained are shoen in the table below:

/* Table to be added */