error occurs when determine data amount

[tfpeach]

I have a very simple task. It need recognize three words. So the training data is very little about 0.15 hour.

When I run "sphinxtrain run". It failed at the stage of "Phase 5: Determine amount of training data, see if n_tied_states seems reasonable." The message shows:
Estimated Total Hours Training: 0.140961111111111
WARNING: Not enough data for the training

I have set $CFG_N_TIED_STATES = 10.

I think 10 is smaller enough. Why can't I pass the check?

Thank you very much!

[Nickolay V. Shmyrev]

Set

# (yes/no) Train context-dependent models
$CFG_CD_TRAIN = 'no';

[tfpeach]

Thank you.

However, it still fail at the step of check the data amount. I attach my log file and configuration file. You can check. Why is it failed? I can not tell any mistake here.

[tfpeach]

I updated sphinxtrain. It can go through the training.

However, when I do the decoding. The WER is 100% (I only use the .ci_cont model to do decoding). All the output is NULL. Please see the attachment which is the recognition result.

I set 4 gaussians. I train 5 iterations after each split (split 1 to 2 gaussians, and 2 to 4 gaussians).

Even though the data set is small, the task is pretty simple. Why do I still get this bad result?

Thank you.

[Sri Ningsih]

I updated sphinxtrain. It can go through the training.

can you explain the detail, what sphinxtrain did you updated ?
i try to train with 15 minutes data training, but i always got failed in phase 5. i used sphinxtrain 1.0.8, sphinxbase 0.8, pocketsphinx 0.8 for training.

[Nickolay V. Shmyrev]

Latest version is 5prealpha

If you want to get help on error you need to share your training folder.

[Sri Ningsih]

thank you nickolay,

this is my folder training

i used 5prealpha for training

[Nickolay V. Shmyrev]

You'd better share the training folder to get help on this issue. Maybe your grammar is not correct or something, it is hard to say.