MFCC doesn't match dimension

[Hamdan Prakoso]

Hi,
I did training with my 2 Hrs data and got these error

MODULE: 90 deleted  interpolation
    Phase 1: Cleaning up directories: logs...
    Phase 2: Doing interpolation...
    Phase 3: Dumping senones for PocketSphinx...
MODULE: DECODE Decoding using models previously trained
        Decoding 1056 segments starting at 0 (part 1 of 1) 
        0% ERROR: FATAL: "ms_gauden.c", line 256: /home/hamdanpe/Documents/data_training/an4/model_parameters/an4.cd_semi_200/variances: #mfcc_ts(6984) doesn't match dimensions: 1 x 8 x 51

ERROR: Failed to start pocketsphinx_batch
        Aligning results to find error rate
        SENTENCE ERROR: 1.5% (16/1056)   WORD ERROR RATE: 0.2% (15/7488)

Have somebody had a similar error ? I found similar thread, but no answer
I don't have any idea where my mistake is... What should I do ?
Thank you

[Nickolay V. Shmyrev]

Based on the message there were issues in training, you need to provide the model training folder in a zip file to get help on this issue.

[Hamdan Prakoso]

Thank you for the answer
Those issues came when I'm using continuous in training configuration. When I changed that to semi-continuous , I got no issues , but WER was higher than continuous one.

I got another issue when using the model to integrate with Visual Studio 2015 . I'm using it to build Windows Phone App using PocketSphinx. I used Toine's project in github as reference.

I put my model_parameters folder and dict file to project and when deploy the project, I got issue saying "Couldn't create decoder". I've attach my model and dict file also the screenshot showing error. Thank you

[Nickolay V. Shmyrev]

I put my model_parameters folder and dict file to project

I see you changed number of gaussians in your model which you need to revert to default recommended by tutorial.

You need to share a whole model training folder, not just a model parameters.

I put my model_parameters folder and dict file to project and when deploy the project, I got issue saying "Couldn't create decoder". I've attach my model and dict file also the screenshot showing error. Thank you

You can check decoder log for details, in Windows it is stored in

char *logPath = concat(applicationLocalStorageFolder, "\\errors.log");

[Hamdan Prakoso]

This is my whole model training folder : https://onedrive.live.com/redir?resid=80BC91E0116D9664!63571&authkey=!AOgSWzFJanKFSsU&ithint=file%2czip

I forgot whether I changed that number of gaussian or not. But what is the recommended number for those number and where I can change it ?

Thank you

[Nickolay V. Shmyrev]

These two should be 256 instead of 8.

if ($CFG_HMM_TYPE eq '.semi.') {
  $CFG_DIRLABEL = 'semi';
# Four stream features for PocketSphinx
  $CFG_FEATURE = "s2_4x";
  $CFG_NUM_STREAMS = 4;
  $CFG_INITIAL_NUM_DENSITIES = 8;
  $CFG_FINAL_NUM_DENSITIES = 8;
  die "For semi continuous models, the initial and final models have the same density".
    if ($CFG_INITIAL_NUM_DENSITIES != $CFG_FINAL_NUM_DENSITIES);

[Hamdan Prakoso]

Hi, Nickolay. Thank you. Already fix that and no error on decoder anymore in my WP project.
But, Error (again) came up when speechRecognizer.AddGrammarSearch method called. I put my .gram there as parameter, but I wasn't sure that my jsgf is correct.

The error said that "fault adding grammar search: Digits".

Errors.log from local storage provided in attachment.

[Nickolay V. Shmyrev]

The error log is incomplete because it exited too early. You can add a call of fflush(stderr) to err.c in sphinxbase in order to flush messages as they printed. Then you'll see a complete error.

[Nickolay V. Shmyrev]

In err_logfp_cb function.