Training Model - (Please advice if the training done is proper)

[Senjam Shantirani]

I have done training using sphinxbase-0.6.

Database: data
Dictionary: CMU (Added few words myself with suitable phonetic sequence)

The trainging resulted in the creating below mentioned folders, each having four files: means, mixture_weights, variances, transition_matrices.

1-datat.falign_ci_cont_1
2-data.falign_ci_cont_2
3- data.falign_ci_cont_4
4- data.falign_ci_cont_8
5- data.falign_ci_cont_16
6- data.falign_ci_cont_32

It was shown in the terminal that the "Training for 32 Gaussian(s) completed after iterations, BUT the training ended with the error below:

Training for 32 Gaussian(s) completed after  iterations
Phase 6: Running force alignment in  1 parts
    Force alignment starting: (1 of 1) 
    0% FATAL_ERROR: "main_align.c", line 1064: fopen(/home/PhoneModel/workspace/hmm/vtlnout/0.80/data.alignedtranscripts.1,r) failed
    Failed to run sphinx3_align
    Failed in part 1
    Something failed: (/home/PhoneModel/workspace/hmm/scripts_pl/04.vtln_align/slave_align

I tested the models in timit.falign_ci_cont_32 with sphinx3_decode and sphinx3_align and gave good results.

My queries are:

A) Since the test results are good from sphinx3_decode and sphinx3_align and models from data.falign_ci_cont_32, Should I consider the training was done properly? Or do I need to do something else?

B) what all parameter I should set in the config file to create a model like wsj_all_cd30.mllt_cd_cont_4000?

I have attached my sphinx_train.cfg and RunAll.pl for your reference.

Thanking you in advance,

Senjam Shantirani

[Nickolay V. Shmyrev]

You need to use latest version according to tutorial

http://cmusphinx.sourceforge.net/wiki/tutorialam

[Senjam Shantirani]

Thank you, I tried this before the reply came and it worked.