Hi,
I want to build decision tree but i got fatal error after the warning said non triphone involved in my base phone file appeared. below is the implementation detail:
It seems to me this is saying you defined a phone named "eset", but the eset phone never occurred in your training data. Or at least, there's no record in your cd_untied_3s.mdef file of eset having appeared in the initial training.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
ooo i see...argument -phone should be filled with one phone only, i thought it want to access to CI phone file like other executable before.
i have nine phones (E-set) and use 3 state.so, i should run the bldtree 9*3=27 times, isn't it? 9 times for state 0, 9 times for state 1 and 9 times for state 2, am i right?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Yes, that is what the Perl script slave.treebuilder.pl does.
I recommed you use the provided Perl scripts. I found it easiest to always invoke RunAll.pl. As I proceeded, I would comment out steps I thought had completed successfully. But be prepared to go back and re-do previous steps when an error manifests itself later. I had to do that many times.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
I want to build decision tree but i got fatal error after the warning said non triphone involved in my base phone file appeared. below is the implementation detail:
bin/bldtree \
-treefn etc/dec_tree.list \
-moddeffn model_architecture/cd_untied_3s.mdef \
-mixwfn model_parameters/cd_mixw_f.pm \
-ts2cbfn .cont. \
-meanfn model_parameters/cd_untied_means_f.pm \
-varfn model_parameters/cd_untied_var_f.pm \
-mwfloor 1e-4 \
-psetfn etc/questions.list \
-phone etc/eset.phone \
-state 0 \
-stwt 1.0 0.3 0.1 \
-ssplitmin 1 \
-ssplitmax 5 \
-ssplitthr 8e-4 \
-csplitmin 1 \
-csplitmax 100 \
-csplitthr 8e-4 \
-cntthresh 0.00001
[Switch] [Default] [Value]
-treefn etc/dec_tree.list
-moddeffn model_architecture/cd_untied_3s.mdef
-ts2cbfn .semi. .cont.
-meanfn model_parameters/cd_untied_means_f.pm
-varfn model_parameters/cd_untied_var_f.pm
-varfloor 0.00001 1.000000e-05
-cntthresh 0.00001 1.000000e-05
-mixwfn model_parameters/cd_mixw_f.pm
-psetfn etc/questions.list
-phone etc/eset.phone
-state 0
-mwfloor 1e-4 1.000000e-04
-stwt 1.0 0.3 0.1
-ssplitthr 8e-4 8.000000e-04
-ssplitmin 1 1
-ssplitmax 5 5
-csplitthr 8e-4 8.000000e-04
-csplitmin 1 1
-csplitmax 100 100
INFO: main.c(183): Reading: model_architecture/cd_untied_3s.mdef
INFO: model_def_io.c(593): Model definition info:
INFO: model_def_io.c(594): 27 total models defined (10 base, 17 tri)
INFO: model_def_io.c(595): 108 total states
INFO: model_def_io.c(596): 81 total tied states
INFO: model_def_io.c(597): 30 total tied CI states
INFO: model_def_io.c(598): 10 total tied transition matrices
INFO: model_def_io.c(599): 4 max state/model
INFO: model_def_io.c(600): 20 min state/model
WARNING: "main.c", line 216: No triphones involving etc/eset.phone
FATAL_ERROR: "ckd_alloc.c", line 85: Calloc failed from main.c(699)
actually, what was mistake? can anyone here help me?thanks...
It seems to me this is saying you defined a phone named "eset", but the eset phone never occurred in your training data. Or at least, there's no record in your cd_untied_3s.mdef file of eset having appeared in the initial training.
ooo i see...argument -phone should be filled with one phone only, i thought it want to access to CI phone file like other executable before.
i have nine phones (E-set) and use 3 state.so, i should run the bldtree 9*3=27 times, isn't it? 9 times for state 0, 9 times for state 1 and 9 times for state 2, am i right?
Yes, that is what the Perl script slave.treebuilder.pl does.
I recommed you use the provided Perl scripts. I found it easiest to always invoke RunAll.pl. As I proceeded, I would comment out steps I thought had completed successfully. But be prepared to go back and re-do previous steps when an error manifests itself later. I had to do that many times.
Still have the test case? Please send a tar ball to me. -Arthur