HMM using Kevin Murphy matlab toolbox

[EDIT: 20110626 10:27 CDT - reformat - WDR]

Hi,

I am facing issues while building a HMM based recognizer for the recognition of 10 different words.

Details:

toolbox -- http://www.cs.ubc.ca/~murphyk/Software/HMM/hmm.html HMM- mixtures of gaussian.

what should be the values of these parameters in the ?? Feature vector (for one word) size 100X39. 100-frames. 39 feature dimension.

O = 8;          %Number of coefficients in a vector 
T = 420;        %Number of vectors in a sequence 
nex = 1;        %Number of sequences 
M = 1;          %Number of mixtures 
Q = 6;          %Number of states

for reference please check the code mhmm_em_demo.m in this toolbox.

10 different words from different speakers. Does this imply we need to have 10 HMM models for each word.

Lets say my feature vector is of 100X39 --What should be the output symbols and can we keep hidden states as any number say 5. Also, how to store this data since this toolbox requires data in three dimensional.

As I have just started on HMM so may be my question look pretty ordinary. Looking forward for a reply. Thanks for the help.

Regards

Prats

[Fred Lawton]

Did you ever get help? I'm trying to use Kevin Murphy's toolbox, and I'm having some problems.

[RITESH]

Prats, Are you done with your HMM based Speech Recognizer? of so, can you share the file with me, i mean the source code for reference. I have no idea where to start and how to develop it from scratch.Also may i know how much time it may involve?