Training for .const works, but .semi and .pmt fail

[Toine db]

I'm getting again and again the same error, even when I try to work with $CFG_INITIAL_NUM_DENSITIES and
$CFG_FINAL_NUM_DENSITIES.

Error at Module 90:

Phase 2: Doing interpolation...
ERROR: This step had 1 ERROR messages and 0 WARNING messages. Please check the log file for details.
completed
Phase 3: Dumping senones for PocketSphinx...
ERROR: FATAL: "senone.c", line 213: fopen(f:/sphinxtraining/neh/model_parameters/neh.cd_semi_200_delinterp/mixture_weights,rb) failed
ERROR: This step had 1 ERROR messages and 0 WARNING messages. Please check the log file for details.

Any suggestion how this could happen?

possibly any relations with this thread https://sourceforge.net/p/cmusphinx/discussion/help/thread/7010d828/ !

Logdir: https://onedrive.live.com/redir?resid=53DF68CA92747BA6!224154&authkey=!ABC6mWqzQZHP_ZE&ithint=file%2crar

[Nickolay V. Shmyrev]

This is expected, there is no much sense to train semi context-independent model.

[Toine db]

OK, got it.

You suggested, a long long time ago, that I use 4 gausians http://bit.ly/1U75E00

I thought you meant in the training, because I couldn't find any attribute for that anywhere?

[Toine db]

Finaly found it how to set the 4 gausians you mentioned earlier.

Adjusted the training config like: and WER got lower after

Single stream features - Sphinx 3
$CFG_FEATURE = "1s_c_d_dd";
$CFG_NUM_STREAMS = 1;
$CFG_INITIAL_NUM_DENSITIES = 1;
$CFG_FINAL_NUM_DENSITIES = 4;
die "The initial has to be less than the final number of densities"
if ($CFG_INITIAL_NUM_DENSITIES > $CFG_FINAL_NUM_DENSITIES);

Thanks for the tip