Training of TIdigits

[Saumya Vij]

Hello!

I am training TIdigits in pocketsphinx. I am getting the following error in
the logdir:( I am pasting the initial and the final part of the kmeans
logdir). Please help if you have any idea how to resolve it.
/CLUSTERHOMES/LMS_AUDIO/maas/vij/REMOS_HTK341_project/ASR_systems/pocketSphinx
/train_clean_wav/bin/kmeans_init \
-gthobj single \
-stride 1 \
-ntrial 1 \
-minratio 0.001 \
-ndensity 256 \
-meanfn /CLUSTERHOMES/LMS_AUDIO/maas/vij/REMOS_HTK341_project/ASR_systems/pocketSphinx/train_clean_wav/model_parameters/train_clean_wav.ci_semi_flatinitial/means \
-varfn /CLUSTERHOMES/LMS_AUDIO/maas/vij/REMOS_HTK341_project/ASR_systems/pocketSphinx/train_clean_wav/model_parameters/train_clean_wav.ci_semi_flatinitial/variances \
-fullvar no \
-reest no \
-segdmpdirs /CLUSTERHOMES/LMS_AUDIO/maas/vij/REMOS_HTK341_project/ASR_systems/pocketSphinx/train_clean_wav/bwaccumdir/train_clean_wav_buff_1 \
-segdmpfn /CLUSTERHOMES/LMS_AUDIO/maas/vij/REMOS_HTK341_project/ASR_systems/pocketSphinx/train_clean_wav/bwaccumdir/train_clean_wav_buff_1/train_clean_wav.dmp \
-ceplen 13 \
-feat s2_4x \
-agc none \
-cmn current \
-varnorm no

Current configuration:

-agc none none
-agcthresh 2.0 2.000000e+00
-cbcntfn
-cepdir
-cepext .mfc .mfc
-ceplen 13 13
-cmn current current
-cmninit 8.0 8.0
-ctlfn
-dictfn
-dmoddeffn
-example no no
-fdictfn
-feat 1s_c_d_dd s2_4x
-featsel 0
-fpcachesz 3000 3000
-fullvar no no
-gthobj state single
-help no no
-lda
-ldadim 0 0
-lsnfn
-maxcbobs 0
-maxiter 100 100
-maxtotobs 0
-meanfn /CLUSTERHOMES/LMS_AUDIO/maas/vij/REMOS_HTK341_project/ASR_systems/pocketSphinx/train_clean_wav/model_parameters/train_clean_wav.ci_semi_flatinitial/means
-method rkm rkm
-minratio 0.01 1.000000e-03
-mixwfn
-ndensity 256
-niter 20 20
-ntrial 5 1
-obscachesz 92 92
-omoddeffn
-reest yes no
-runlen 0
-segdir
-segdmpdirs /CLUSTERHOMES/LMS_AUDIO/maas/vij/REMOS_HTK341_project/ASR_systems/pocketSphinx/train_clean_wav/bwaccumdir/train_clean_wav_buff_1,
-segdmpfn /CLUSTERHOMES/LMS_AUDIO/maas/vij/REMOS_HTK341_project/ASR_systems/pocketSphinx/train_clean_wav/bwaccumdir/train_clean_wav_buff_1/train_clean_wav.dmp
-segext v8_seg v8_seg
-segidxfn
-stride 32 1
-svspec
-ts2cbfn
-tscnt 0
-tsoff 0 0
-tsrngfn
-varfn /CLUSTERHOMES/LMS_AUDIO/maas/vij/REMOS_HTK341_project/ASR_systems/pocketSphinx/train_clean_wav/model_parameters/train_clean_wav.ci_semi_flatinitial/variances
-varnorm no no
-vartiethr 0 0

INFO: feat.c(684): Initializing feature stream to type: 's2_4x', ceplen=13,
CMN='current', VARNORM='no', AGC='none'
INFO: cmn.c(142): mean= 12.00, mean= 0.0
INFO: main.c(519): No mdef files. Assuming 1-class init
INFO: main.c(1351): 1-class dump file
INFO: main.c(1389): Corpus 0: sz==1306112 frames
INFO: main.c(1398): Convergence ratios are abs(cur - prior) / abs(prior)
INFO: main.c(231): alloc'ing 59Mb obs buf
INFO: main.c(576): Initializing means using random k-means
INFO: main.c(579): Trial 0: 256 means
INFO: kmeans.c(153): km iter 1.000000e+00 ...
WARNING: "kmeans.c", line 431: Empty cluster 2
WARNING: "kmeans.c", line 431: Empty cluster 3
WARNING: "kmeans.c", line 431: Empty cluster 4
WARNING: "kmeans.c", line 431: Empty cluster 5
WARNING: "kmeans.c", line 431: Empty cluster 6
WARNING: "kmeans.c", line 431: Empty cluster 7
WARNING: "kmeans.c", line 431: Empty cluster 8
WARNING: "kmeans.c", line 431: Empty cluster 9
WARNING: "kmeans.c", line 431: Empty cluster 10
WARNING: "kmeans.c", line 431: Empty cluster 11
WARNING: "kmeans.c", line 431: Empty cluster 12
.
.
.
WARNING: "kmeans.c", line 431: Empty cluster 234
WARNING: "kmeans.c", line 431: Empty cluster 235
WARNING: "kmeans.c", line 431: Empty cluster 236
WARNING: "kmeans.c", line 431: Empty cluster 237
WARNING: "kmeans.c", line 431: Empty cluster 238
WARNING: "kmeans.c", line 431: Empty cluster 239
WARNING: "kmeans.c", line 431: Empty cluster 240
WARNING: "kmeans.c", line 431: Empty cluster 241
WARNING: "kmeans.c", line 431: Empty cluster 242
WARNING: "kmeans.c", line 431: Empty cluster 243
WARNING: "kmeans.c", line 431: Empty cluster 244
WARNING: "kmeans.c", line 431: Empty cluster 245
WARNING: "kmeans.c", line 431: Empty cluster 246
WARNING: "kmeans.c", line 431: Empty cluster 247
WARNING: "kmeans.c", line 431: Empty cluster 248
WARNING: "kmeans.c", line 431: Empty cluster 249
WARNING: "kmeans.c", line 431: Empty cluster 250
WARNING: "kmeans.c", line 431: Empty cluster 251
WARNING: "kmeans.c", line 431: Empty cluster 252
WARNING: "kmeans.c", line 431: Empty cluster 253
WARNING: "kmeans.c", line 431: Empty cluster 254
WARNING: "kmeans.c", line 431: Empty cluster 255
INFO: main.c(612): -> Aborting k-means, bad initialization
INFO: main.c(621): best-so-far sqerr = -1.000000e+00
ERROR: "main.c", line 840: Too few observations for kmeans
ERROR: "main.c", line 1407: Unable to do k-means for state 0; skipping...
INFO: s3gau_io.c(226): Wrote /CLUSTERHOMES/LMS_AUDIO/maas/vij/REMOS_HTK341_pro
ject/ASR_systems/pocketSphinx/train_clean_wav/model_parameters/train_clean_wav
.ci_semi_flatinitial/means
INFO: s3gau_io.c(226): Wrote /CLUSTERHOMES/LMS_AUDIO/maas/vij/REMOS_HTK341_pro
ject/ASR_systems/pocketSphinx/train_clean_wav/model_parameters/train_clean_wav
.ci_semi_flatinitial/variances
INFO: main.c(1508): No mixing weight file given; none written
INFO: main.c(1668): TOTALS: km 0.009x 1.019e+00 var 0.000x 0.000e+00 em 0.000x
0.000e+00 all 0.009x 1.019e+00

[Saumya Vij]

Please help me if somebody has an idea about this problem.

[Nickolay V. Shmyrev]

Error says that you do not have enough data for training:

ERROR: "main.c", line 840: Too few observations for kmeans

[Saumya Vij]

But I do have sufficient data for training as I have trained it before.

[Nickolay V. Shmyrev]

But I do have sufficient data for training as I have trained it before.

Believe me if trainer message gives you some error it's usually indeed the
situation. Software rarely makes mistakes unless its programmed to do them.

You need to search for the issue in the things you have done, not in the
trainer itself. You did something wrong. Check everything, check the features
extracted, maybe they were not extracted properly. Check logs from logdir for
for other tools invoked. They might give you more information.

You always have an option to share your training folder and logs. That will
help you to get the answer faster.

[Nickolay V. Shmyrev]

According to the log, features might not have enough diversity. All 1306112
frames might have zero values. That will not allow to perform clustering with
kmeans.

You can dump mfc files with sphinx_cepview command.