I have recently started using HTK. There was this peculiar behavior that I =
 noticed about HCompV. I am not sure whether the issue has been addressed b=
efore. I couldnt find it in the archives.

The issue is that the parameter values generated by HCompV show sensitivity=
 to the order in which the input files, to be used for computing the parame=
ters, are listed in .scp file. The sensitivity has been enough to affect my=
 speech recognizer results.

HCompV -m -S train_mfcc_paths.scp  -M hmm1.a  proto.txt

This how i have been using the HCompV tool.=20
A sample of the .scp file is as follows:

.\mfc\f11aba1.mfcc
.\mfc\f11aba2.mfcc
.\mfc\f11abi1.mfcc
.\mfc\f11abi2.mfcc
.\mfc\f11abu1.mfcc
.\mfc\f11abu2.mfcc
.\mfc\f11acha1.mfcc
.\mfc\f11acha2.mfcc
.\mfc\f11achi1.mfcc
.\mfc\f11achi2.mfcc
.\mfc\f11achu1.mfcc
.\mfc\f11achu2.mfcc
.\mfc\f11ada1.mfcc
.\mfc\f11ada2.mfcc
.\mfc\f11adha1.mfcc
.\mfc\f11adha2.mfcc
.\mfc\f11adhi1.mfcc
.\mfc\f11adhi2.mfcc
.\mfc\f11adhu1.mfcc
.\mfc\f11adhu2.mfcc
.\mfc\f11adi1.mfcc
.\mfc\f11adi2.mfcc
.\mfc\f11adja1.mfcc
.\mfc\f11adja2.mfcc
.\mfc\f11adji1.mfcc
.\mfc\f11adji2.mfcc
.\mfc\f11adju1.mfcc
.\mfc\f11adju2.mfcc
.\mfc\f11adu1.mfcc
.\mfc\f11adu2.mfcc

The  order in which the above files are listed in the .scp file is affectin=
g the parameters computed.

I have been using HTK 3.4 with Windows XP.

My config file used by HCopy for generating the feature vectors is as follo=
ws:

# coding parameters
SOURCEFORMAT =3D WAV
TARGETKIND =3D MFCC_D_A_Z_0
TARGETRATE =3D 100000.0
SAVECOMPRESSED =3D F
SAVEWITHCRC =3D F
WINDOWSIZE =3D 250000.0
USEHAMMING =3D T
PREEMCOEF =3D 0.97
NUMCHANS =3D 26
CEPLIFTER =3D 22
NUMCEPS =3D 12
ENORMALISE =3D F

I would be glad if somebody confirmed this behavior and explained it.