I'm going to implement MMIE using SphinxTrain. However, I'm a newcomer and I only know something about the process of training: I've read the Training doc and tried to run the tutorial. I'm trying to read the source code and implement the MMIE algorithm by Povey(Cambridge). But so many related files and complex codes really frustrate me. Who can give me some advice or tell me the one-by-one steps? Thanks~ Anything is welcome.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
All baum welch algorithm implementation are not easy to understand. Reading the code is still the best way to start.
Plus you are dealing with MMIE which is one level more difficult than ML estimation. So I don't feel surprised you would find it a little bit difficult.
I recommend you to talk to Prof. Yannick Esteve (http://www-lium.univ-lemans.fr/~esteve/research.html), he has a lot of experience in SphinxTrain's hacking, especially AM. He might be able to give you some hacking advice.
Arthur
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
A still good reference for Baum Welch reestimation is Huang, X. D. and M. A. Jack. 1989. Semi-Contiunous Hidden Markov Model for Speech Recognition. Computer Speech and Language, 3(3).
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I'm going to implement MMIE using SphinxTrain. However, I'm a newcomer and I only know something about the process of training: I've read the Training doc and tried to run the tutorial. I'm trying to read the source code and implement the MMIE algorithm by Povey(Cambridge). But so many related files and complex codes really frustrate me. Who can give me some advice or tell me the one-by-one steps? Thanks~ Anything is welcome.
All baum welch algorithm implementation are not easy to understand. Reading the code is still the best way to start.
Plus you are dealing with MMIE which is one level more difficult than ML estimation. So I don't feel surprised you would find it a little bit difficult.
I recommend you to talk to Prof. Yannick Esteve (http://www-lium.univ-lemans.fr/~esteve/research.html), he has a lot of experience in SphinxTrain's hacking, especially AM. He might be able to give you some hacking advice.
Arthur
A still good reference for Baum Welch reestimation is Huang, X. D. and M. A. Jack. 1989. Semi-Contiunous Hidden Markov Model for Speech Recognition. Computer Speech and Language, 3(3).