I've trained a set of fully continuous acoustic model using the automation script of the SphinxTrain module provided at www-2.cs.cmu.edu/~robust/Tutorial/ (tutorial_single_machine.tar.gz) and also decoded the test utterances using the Sphinx3 decoder module provided at the same location. Both the modules seem to work out fine for me (i.e. the test accuracy seems to be satisfactory).
However, when I tried to load the same set of acoustic model (8 gaus_cd_continuous HMM model) to my Sphinx3.3 decoder, the recognition performance tends to be extremely poor. I've checked to ensure that i use the proper set of LM DMP file, word dictionary etc during decoding. What could've been the problem? Could it be different version of the decoder used (i.e. Sphinx3 vs Sphinx3.3).
Thanks in advance for any help.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I've trained a set of fully continuous acoustic model using the automation script of the SphinxTrain module provided at www-2.cs.cmu.edu/~robust/Tutorial/ (tutorial_single_machine.tar.gz) and also decoded the test utterances using the Sphinx3 decoder module provided at the same location. Both the modules seem to work out fine for me (i.e. the test accuracy seems to be satisfactory).
However, when I tried to load the same set of acoustic model (8 gaus_cd_continuous HMM model) to my Sphinx3.3 decoder, the recognition performance tends to be extremely poor. I've checked to ensure that i use the proper set of LM DMP file, word dictionary etc during decoding. What could've been the problem? Could it be different version of the decoder used (i.e. Sphinx3 vs Sphinx3.3).
Thanks in advance for any help.