Here is another update. I used class based LM with Sphinx4, and got pretty good recognition.
This points me back to some bug in sphinx3 integration of class based LM.
Can anyone please help. I am happy to share the LM and the dictionary.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I used class based LM in Sphinx3 with good results -not full-class model-. You can obtain an example of use in Sphinx2 package. If you can't found the example write me.
Alejandro Renato
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hello arenato,
I am using a LM with 60K work vocabulary with class and within class word probabilities.
My problem is that Sphinx3 seems to be ignoring these probabilities. So the recognized sentance
actually has a very high perplexity because recognizer basically ignored these probs.
The turtle LM does not have probabilities and is too small to show this effect. What size is your LM?
Nagendra
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Here is another update. I used class based LM with Sphinx4, and got pretty good recognition.
This points me back to some bug in sphinx3 integration of class based LM.
Can anyone please help. I am happy to share the LM and the dictionary.
I used class based LM in Sphinx3 with good results -not full-class model-. You can obtain an example of use in Sphinx2 package. If you can't found the example write me.
Alejandro Renato
Hello arenato,
I am using a LM with 60K work vocabulary with class and within class word probabilities.
My problem is that Sphinx3 seems to be ignoring these probabilities. So the recognized sentance
actually has a very high perplexity because recognizer basically ignored these probs.
The turtle LM does not have probabilities and is too small to show this effect. What size is your LM?
Nagendra