Hi all,

When i run my HTK recognizer live without using julius i get a language model score of 0.0.

I have followed the tutorial on voxforge on how to build htk and only changed the mean and variance and the type for my hmm (39,39,MFCC_0_D). Is there something wrong with my system? It shouldn't be 0 because i am using a language model which is the wdnet(I may be wrong, please give me more info if you can. I am quite new to the concept)

I would like to get the language model score so that maybe i can use it to compute the confidence score for rejecting outputs that are below the threshold.

The reason that i would like to use htk alone and not with julius is because julius is slow to recognize when compared to htk. There is also a problem of it not able to catch my speech signal from a distance when htk can(using the exact same microphone). Does anybody else have this problem?

Thank you

--- (Edited on 7/30/2015 5:24 am [GMT-0500] by Visitor) ---