Hi, in PocketSphinx, the number of n_tied_state in the mdef file is diffrent and much bigger than the corresponding number at the beginning of the definitions of each means and variances files. Does anyone know how are those vectors (I guess reused) in the different Hmm. I couldn't find anything about that.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I know it's not very intelligent, but it's the purpose of my master thesis. I wanted to know the algorithme behind extracting this kind of Hmm (see picture 1 below) from Pocketsphinx files. I used the universal speech converter that's from 2005 and it worked succefully with the sphinx5.2 prealpha continuous. But with semi-continuous it doesn't work properly. I suspect that it may also be because of the multi-streaming. On picture 2 you can see the number of n_tied_state in both mdef and means (debinarised) they are the same in continuous model but very different in pocketsphinx.
Hi, in PocketSphinx, the number of n_tied_state in the mdef file is diffrent and much bigger than the corresponding number at the beginning of the definitions of each means and variances files. Does anyone know how are those vectors (I guess reused) in the different Hmm. I couldn't find anything about that.
It is not quite clear what are you asking, you need to be more specific about numbers you mean
Acoustic model format is covered in our documentation:
https://cmusphinx.github.io/wiki/acousticmodelformat/
I know it's not very intelligent, but it's the purpose of my master thesis. I wanted to know the algorithme behind extracting this kind of Hmm (see picture 1 below) from Pocketsphinx files. I used the universal speech converter that's from 2005 and it worked succefully with the sphinx5.2 prealpha continuous. But with semi-continuous it doesn't work properly. I suspect that it may also be because of the multi-streaming. On picture 2 you can see the number of n_tied_state in both mdef and means (debinarised) they are the same in continuous model but very different in pocketsphinx.
You can read about semi-continuous model structure here:
https://cmusphinx.github.io/wiki/acousticmodeltypes/