i establish recognition system based in mffcc isolated word(i followed HTK 3.4 tutorial part 3 untill step 11 ). I perform the whole recognition without get error. But when itry to test the recognition system, I only get 18% of recognition rate. I think that i didn't make a good choice of training data then my question is how arranging the data base for recognition and test step. I used a clean Aurora data base(with nohead data). My config file is as follow:    TRAGETKIND=MFCCC_0_D_A

SOURCEFORMAT=NOHEAD

TRAGETRATE=100000.0

HNET:TRACE=1

SAVECOMPRESSED=T

SAVEWITHCRC=T

USEHAMMING=T

ZMEANSOURCE=F

WINDOWSIZE=250000.0

ENORMALISE=F

PREEMCOEF=0.97

CEPLIFTER=22

NUMCHANS=26

NUMCEPS=12

SOURCERATE=1250.0

SAVEBINARY=F

HMATH:USELOGADDTABLE=F

--- (Edited on 5/13/2011 2:09 pm [GMT-0500] by raf23) ---