This experiment was on clean train and test data. Training was done on 7138 utterences while testing was done on devel set with 330 utterences.
The baseline performance was 10.2 % with isip front end.
The same experiment was run on cluster as well as on servers to study the variation in the WER with increase in the number of processors.
The experiment was run for 16, 8 and 4 mixtures. As expected the WER for 4 mixture model was higher and improved significantly for 8 mixture model.

The results for the above experiment can be found here:

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

/cavs/hse/ies/isip/data/research/nonlinear/exp_011/aurora/asr_wsj_tutorial_v1.4.2/exp_1