mfcc doesn't match dimension

[Edwin Miguel Triana]

Hi People,
Currently I'm building acoustic models (AM) for Brazilian Portugues for
pocketsphinx, during the training process I get some errors but at the end I
get the AMs, then in the decoding phase I get the following error using
pocketsphinx:

INFO: cmn.c(142): mean[0]= 12.00, mean[1..12]= 0.0
INFO: mdef.c(520): Reading model definition: /home/etriana/speech/exp/exp0022/model_parameters/exp0022.cd_semi_1000/mdef
INFO: bin_mdef.c(173): Allocating 22868 * 8 bytes (178 KiB) for CD tree
INFO: tmat.c(205): Reading HMM transition probability matrices: /home/etriana/speech/exp/exp0022/model_parameters/exp0022.cd_semi_1000/transition_matrices
INFO: acmod.c(117): Attempting to use SCHMM computation module
INFO: ms_gauden.c(198): Reading mixture gaussian parameter: /home/etriana/speech/exp/exp0022/model_parameters/exp0022.cd_semi_1000/means
INFO: ms_gauden.c(292): 1 codebook, 4 feature, size: INFO: ms_gauden.c(294):  256x12INFO: ms_gauden.c(294):  256x24INFO: ms_gauden.c(294):  256x3INFO: ms_gauden.c(294):  256x12INFO: ms_gauden.c(295): 
INFO: ms_gauden.c(198): Reading mixture gaussian parameter: /home/etriana/speech/exp/exp0022/model_parameters/exp0022.cd_semi_1000/variances
FATAL_ERROR: "ms_gauden.c", line 254: /home/etriana/speech/exp/exp0022/model_parameters/exp0022.cd_semi_1000/variances: #mfcc_ts(223488) doesn't match dimensions: 1 x 256 x 51

That is the error from logdir/decode/exp0022-1-1.log . I just modified the
/home/etriana/speech/sphinx/pocketsphinx/src/libpocketsphinx/mdef.c to be
case sensitive, a requiremnt for SAMPA.

Have somebody had a similar error? Could somebody give a hint to solve this
problem?

Thanks in advance for your help!.

Miguel

[Nickolay V. Shmyrev]

I suggest you to share your training folder

[Edwin Miguel Triana]

Hi Nickolay,
Do you want I share the training folder through http or something like that?
If so, how is the best option to share this?
Thanks

[Nickolay V. Shmyrev]

Public sharing hosting like rapidshare or mediafire should work best.

[Mikhail Barg]

I have a similar problem:

Training for 16 Gaussian(s) completed after 6 iterations
MODULE: 60 Lattice Generation
Skipped:  $ST::CFG_MMIE set to 'no' in sphinx_train.cfg
MODULE: 61 Lattice Pruning
Skipped:  $ST::CFG_MMIE set to 'no' in sphinx_train.cfg
MODULE: 62 Lattice Format Conversion
Skipped:  $ST::CFG_MMIE set to 'no' in sphinx_train.cfg
MODULE: 65 MMIE Training
Skipped:  $ST::CFG_MMIE set to 'no' in sphinx_train.cfg
MODULE: 90 deleted interpolation
Skipped for continuous models
MODULE: DECODE Decoding using models previously trained
        Decoding 1973 segments starting at 0 (part 1 of 1)
        0% ERROR: FATAL: "ms_gauden.c", line 256: c:/Dev/SpeechRecognizer/data/acoustic_model/model_parameters/model.cd_cont_4000/variances: #mfcc_ts(103793040) doesn't match dimensions: 4265 x 16 x 39

ERROR: Failed to start pocketsphinx_batch
        Aligning results to find error rate
Can't open c:/Dev/SpeechRecognizer/data/acoustic_model/result/model-1-1.match
word_align.pl failed with error code 65280 at C:\Dev\SpeechRecognizer\tools\sphinxtrain\scripts\decode\slave.pl line 173.

This started after I enabled $CFG_FULLVAR and $CFG_DIAGFULL in sphinx_train.cfg
I'm using SphinxTrain 5prealpha (win32 Release build) under Win 8.1.

[Nickolay V. Shmyrev]

I suggest you to share your training folder