Error under training

[Andreas Ravndal]

ok, so I get this error when training:
MODULE: 90 deleted interpolation (2016-01-13 15:54)
Phase 1: Cleaning up directories: logs...
Phase 2: Doing interpolation...
delint Log File
ERROR: This step had 1 ERROR messages and 0 WARNING messages. Please check the log file for details.
completed

This is what my log file shows:
ERROR: "model_def_io.c", line 413: Unable to open /home/andreas/Documents/Taledatabase/model_architecture/try1.200.mdef for reading: No such file or directory

When I look in this folder myself I see a file named try1.200.ci.mdef, so I think I have something set up wrong in my cfg file. I can't figure out why it's looking for try1.200.mdef and not try1.200.ci.mdef.

[Nickolay V. Shmyrev]

The message says "Please check the log file for details.". Please check them or provide logdir folder. You can pack it into archive.

[Andreas Ravndal]

I checked the log file for details:
ERROR: "model_def_io.c", line 413: Unable to open /home/andreas/Documents/Taledatabase/model_architecture/try1.200.mdef for reading: No such file or directory...
My problem is that I don understand why it is looking for try1.200.mdef and not try1.200.ci.mdef.
The logdir and cfg file is in the post below. I would be grateful if you took a look :)

[Andreas Ravndal]

Here is my cfg file and logdir

[Nickolay V. Shmyrev]

Since you disabled cd training you need to change

$DEC_CFG_MODEL_NAME = "$CFG_EXPTNAME.cd_${CFG_DIRLABEL}_${CFG_N_TIED_STATES}";

to

$DEC_CFG_MODEL_NAME = "$CFG_EXPTNAME.ci_${CFG_DIRLABEL}";

I recommend you to enable CD training.

[Andreas Ravndal]

Thanks! May I ask why I should use context-depentend and not context-independent models? What is the difference between these two?

[Andreas Ravndal]

So I tried what you said and I still get the same error:
**Unable to open /home/andreas/Documents/Taledatabase/model_architecture/try1.200.mdef for reading: No such file or directory

Do you have any other suggestion to what might be wrong?

[Nickolay V. Shmyrev]

I see you are training semi-continuous context-independent model. Train continuous model, revert to default values of configuration variables.

[Andreas Ravndal]

I have now tried training a continous model with default values(except 16khz because I'm using 8khz sample rate on my audiofiles) and I get this error:

MODULE: 20 Training Context Independent models
Phase 1: Cleaning up directories:
accumulator...logs...qmanager...models...
Phase 2: Flat initialize
Phase 3: Forward-Backward
ERROR: Training failed in iteration 1.

which I don't understand starts at all, because the context-dependent model is set to yes and independent to no.
Here is the error that is written in the logfile:

ERROR: "main.c", line 421: # of codebooks in mean/var files, 186, inconsistent with ts2cb mapping 174
FATAL: "main.c", line 1846: initialization failed
Fri Jan 15 17:14:29 2016

[Nickolay V. Shmyrev]

The following phones are duplicated in your phone file, you need to remove duplicates.

K
UH0
UH1
UH2

[Andreas Ravndal]

wow, how could I have missed out on that one... Thanks for your help!
But a new error came up under the decoding part:

ERROR: "ngram_search.c", line 206: Language model/set does not contain </s>, recognition will fail

Why does it search for </s> in my language model? Does this mean that there may be something wrong with my transcription file? Or do I have to use <s> and </s> in the corpus that i train my language model on?

[Nickolay V. Shmyrev]

Why does it search for in my language model?

It is mandatory

Or do I have to use and in the corpus that i train my language model on?

Yes