Hi
I have a small database, consists of only 12 words and have training data about 0.3 hours. The training results works fine and can achieve high accuracy (SER <= 2%) during recognition with external data set (data that is not in training set).
I have train few set of models with different no of tied states and it gives me strange result. The result is bad for 100 tied state, then increase with 500 tied state, then drop again when increasing to 2000 and then increase again when reaching 5000 and 7000.
Is there any rules for setting the no of tied states to produce a good HMM model? Thanks.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi
I have a small database, consists of only 12 words and have training data about 0.3 hours. The training results works fine and can achieve high accuracy (SER <= 2%) during recognition with external data set (data that is not in training set).
I have train few set of models with different no of tied states and it gives me strange result. The result is bad for 100 tied state, then increase with 500 tied state, then drop again when increasing to 2000 and then increase again when reaching 5000 and 7000.
Is there any rules for setting the no of tied states to produce a good HMM model? Thanks.