I have a failure on the MMIE Training. The SphinxTrain revision is 11302
The 1st and 2nd parts of the 65 mmie train has been finished successfully,
without ERRORs, with WARNINGS (about 1% of the total utterances number).
On the 3rd part (normalization) process failed with the single log message in
the norm.log:
Baum welch ran successfully for only 0 utterance! Aborting..
Looking into 65.mmie_train/norm_and_launchbw.pl line 80 I have found, that the
problem is
in the incorrect $n_done calculation.
It seems, that epected log file structure is not consistent with the actual
one.
This is the end of my .bw.denlat.log:
========================================
utt> 65505 192.168.0.116_12e57641304 263 0 341 1.350309e-13
utt> 65506 192.168.0.116_12e576513cc 289 0 211 -1.819378e-13
utt> 65507 192.168.0.116_12e57655450 289 0 1034 8.373857e-14
INFO: main.c(1248): arc_487 is ignored (viterbi run failed)
utt> 65508 192.168.0.116_12e57658489 212 0 495 2.917111e-14
utt> 65509 192.168.0.116_12e5765d038 212 0 207 -2.484124e-14
utt> 65510 192.168.0.116_12e5766d5d3 263 0 301 1.269540e-13
utt> 65511 192.168.0.116_12e5767133a 263 0 663 1.371125e-14
utt> 65512 192.168.0.116_12e57674ce8 289 0 373 -4.368728e-14
utt> 65513 192.168.0.116_12e57678aad 238 0 221 -1.762757e-13
utt> 65514 192.168.0.116_12e57687500 238 0 220 4.582446e-14
overall> 65515 (-0) -1.068189e+01 -6.998243e+05
INFO: s3gau_io.c(478): Wrote /root/yuri/sphinx_recipe/de8KHz/de8KHzS/bwaccumdi
r/deS_buff_1/denlat_gauden_counts with means with vars
INFO: main.c(1872): Counts saved to
/root/yuri/sphinx_recipe/de8KHz/de8KHzS/bwaccumdir/deS_buff_1
Tue Jan 10 02:18:34 2012
===========================================
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I have a failure on the MMIE Training. The SphinxTrain revision is 11302
The 1st and 2nd parts of the 65 mmie train has been finished successfully,
without ERRORs, with WARNINGS (about 1% of the total utterances number).
On the 3rd part (normalization) process failed with the single log message in
the norm.log:
Baum welch ran successfully for only 0 utterance! Aborting..
Looking into 65.mmie_train/norm_and_launchbw.pl line 80 I have found, that the
problem is
in the incorrect $n_done calculation.
It seems, that epected log file structure is not consistent with the actual
one.
This is the end of my .bw.denlat.log:
========================================
utt> 65505 192.168.0.116_12e57641304 263 0 341 1.350309e-13
utt> 65506 192.168.0.116_12e576513cc 289 0 211 -1.819378e-13
utt> 65507 192.168.0.116_12e57655450 289 0 1034 8.373857e-14
INFO: main.c(1248): arc_487 is ignored (viterbi run failed)
utt> 65508 192.168.0.116_12e57658489 212 0 495 2.917111e-14
utt> 65509 192.168.0.116_12e5765d038 212 0 207 -2.484124e-14
utt> 65510 192.168.0.116_12e5766d5d3 263 0 301 1.269540e-13
utt> 65511 192.168.0.116_12e5767133a 263 0 663 1.371125e-14
utt> 65512 192.168.0.116_12e57674ce8 289 0 373 -4.368728e-14
utt> 65513 192.168.0.116_12e57678aad 238 0 221 -1.762757e-13
utt> 65514 192.168.0.116_12e57687500 238 0 220 4.582446e-14
overall> 65515 (-0) -1.068189e+01 -6.998243e+05
INFO: s3gau_io.c(478): Wrote /root/yuri/sphinx_recipe/de8KHz/de8KHzS/bwaccumdi
r/deS_buff_1/denlat_gauden_counts with means with vars
INFO: main.c(1872): Counts saved to
/root/yuri/sphinx_recipe/de8KHz/de8KHzS/bwaccumdir/deS_buff_1
Tue Jan 10 02:18:34 2012
===========================================
Hello
This issue was fixed in revision r11304. Its a small incompatibility between
bw output format and perl parser.
As for MMIE, I can't confirm it improves accuracy still
Thanks! I will try it.