PTM ac model training: FATAL: "main.c", line 339: src mean n_cb (== 40) is inconsistent with ts2cb mapping 1. Most probably phoneset has duplicated phones

[Orest]

I am trying to train a PTM model by following the tutorial, but I get an error on module 20 phase 2:

FATAL: "main.c", line 339: src mean n_cb (== 40) is inconsistent with ts2cb mapping 1. Most probably phoneset has duplicated phones

when I go to logdir/20.ci_hmm/ I can see :

500000ForPTM_15_April.duplicate_codebook.log

which shows the fatal error:

INFO: main.c(261): Reading src /home/500000ForPTM_15_April/model_architecture/500000ForPTM_15_April.ci.mdef
INFO: model_def_io.c(573): Model definition info:
INFO: model_def_io.c(574): 40 total models defined (40 base, 0 tri)
INFO: model_def_io.c(575): 240 total states
INFO: model_def_io.c(576): 200 total tied states
INFO: model_def_io.c(577): 200 total tied CI states
INFO: model_def_io.c(578): 40 total tied transition matrices
INFO: model_def_io.c(579): 6 max state/model
INFO: model_def_io.c(580): 6 min state/model
INFO: main.c(272): Generating semi-continous ts2cb mappingINFO: main.c(299): Reading src /home/500000ForPTM_15_April/model_parameters/500000ForPTM_15_April.ci_ptm_f
latinitial/mixture_weights
INFO: s3mixw_io.c(117): Read /home/500000ForPTM_15_April/model_parameters/500000ForPTM_15_April.ci_ptm_fla
tinitial/mixture_weights [200x4x256 array]
INFO: main.c(309): Reading src /home/500000ForPTM_15_April/model_parameters/500000ForPTM_15_April.ci_ptm_flatinitial/transition_matrices
INFO: s3tmat_io.c(118): Read /home/500000ForPTM_15_April/model_parameters/500000ForPTM_15_April.ci_ptm_flatinitial/transition_matrices [40x5x6 array]
INFO: main.c(318): Reading src /home/500000ForPTM_15_April/model_parameters/500000ForPTM_15_April.ci_ptm_flatinitial/means
INFO: s3gau_io.c(169): Read /home/500000ForPTM_15_April/model_parameters/500000ForPTM_15_April.ci_ptm_flatinitial/means [40x4x256 array]
FATAL: "main.c", line 339: src mean n_cb (== 40) is inconsistent with ts2cb mapping 1. Most probably phoneset has duplicated phones
Sat Apr 18 23:45:45 2015

This is my training print on screen:

http://pastebin.com/nDvdPWuj

This is my configuration file:

http://pastebin.com/seAdccx4

This is my phoneset

http://pastebin.com/zd6MDaty

Training continuous models doesn't bring any issues, so I thought that my problem might be in the configuration file:

The tutorial says: If you are training semi-continuous or PTM model, use 256 gaussians.

  $CFG_INITIAL_NUM_DENSITIES = 256;
  $CFG_FINAL_NUM_DENSITIES = 256;

Should I keep it like this, or I should have instead:

  $CFG_INITIAL_NUM_DENSITIES = 1;
  $CFG_FINAL_NUM_DENSITIES = 256;

[Nickolay V. Shmyrev]

FATAL: "main.c", line 339: src mean n_cb (== 40) is inconsistent with ts2cb mapping 1. Most probably phoneset has duplicated phones

Set CFG_FALIGN_CI_MGAU to no

Should I keep it like this, or I should have instead:

You should keep the original values.

[Orest]

Set CFG_FALIGN_CI_MGAU to no

thanks, it works