VoxForge
VoxForge
hello Mr.Kmacean
I will like help me solve my problem.
I
work in recognition of continuous speech for the first time, I find it
difficult because it is not the same thing with the recognition of
isolated words.
a step of recording and parameterization of sentences, I have done easily. but the problem in the learning phase, exactly in the order Hinit to initialize the HMM models;
HInit-A-D-T 1-S-trainlist.txt model/hmm0 M-H model / proto-hmm_je.txt the I-L data / train / lab je
je: the first word in my sentence.
I wanted to find the model of each word then the recognition of sentence.
is that fair?
--- (Edited on 11/3/2010 10:22 am [GMT-0500] by mawahib) ---
>HInit-A-D-T 1-S-trainlist.txt model/hmm0 M-H model / proto-hmm_je.txt the I-L data / train / lab je
Sorry, we use a flat start approach to model initialization (see step 6 - creating flat start monophones) using HCompV rather than using HInit.
YOu might look at the HTK Book or email archives for more info on HInit
--- (Edited on 11/4/2010 3:29 pm [GMT-0400] by kmaclean) ---