Hi,
i have reached to cd-tied model training stage. i just want a confirmation whether my training is right or wrong. in order to have 8 mixture gaussian in each state, i train cd-tied model in 4 times (init_mixw, bw iteratively, norm, and inc_cmp). but inc_cmp was ran 3 times only because the last step to get the parameter files is norm.
as i know from the sphinxtrain instruction, the last output files from the training are normalized parameter files (mixw,tmat,means,variances) after it get 8 mixture of gaussian (say) for each state, is it correct?
so, i want to know, after cd-tied model training, what process should i do before i can test the performance in the decoder?
thanks.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
i have reached to cd-tied model training stage. i just want a confirmation whether my training is right or wrong. in order to have 8 mixture gaussian in each state, i train cd-tied model in 4 times (init_mixw, bw iteratively, norm, and inc_cmp). but inc_cmp was ran 3 times only because the last step to get the parameter files is norm.
as i know from the sphinxtrain instruction, the last output files from the training are normalized parameter files (mixw,tmat,means,variances) after it get 8 mixture of gaussian (say) for each state, is it correct?
so, i want to know, after cd-tied model training, what process should i do before i can test the performance in the decoder?
thanks.