On Mon, Jun 8, 2009 at 9:13 PM, Tomasz Kuczmarski
<faqster@xxxxxxxxx> wrote:
During the tree based context clustering I get the following error:
=======================================================================
Start tree-based context clustering at nie, 7 cze 2009, 21:33:02 CEST
=======================================================================
/usr/local/HTS-2.1/bin/HHEd -A -B -C /home/tomek/Speech/HTS-demo_AMU_PL_ILO/configs/trn.cnf -D -T 1 -p -i -H /home/tomek/Speech/HTS-demo_AMU_PL_ILO/models/qst001/ver1/cmp/clustered.mmf -m -a 1.0 -w /home/tomek/Speech/HTS-demo_AMU_PL_ILO/models/qst001/ver1/cmp/clustered.mmf /home/tomek/Speech/HTS-demo_AMU_PL_ILO/edfiles/qst001/ver1/cmp/cxc_mgc.hed /home/tomek/Speech/HTS-demo_AMU_PL_ILO/data/lists/full.list
HTK Configuration Parameters[9]
Module/Tool Parameter Value
# MINDUR 5
# MAXSTDDEVCOEF 10
# APPLYDURVARFLOOR TRUE
# DURVARFLOORPERCENTILE 1.000000
# VFLOORSCALESTR Vector 4 0.01 0.01 0.01 0.01
# MINLEAFOCC 10.000000
# NATURALWRITEORDER TRUE
# NATURALREADORDER TRUE
# APPLYVFLOOR TRUE
HHEd
96182/96182 Models Loaded [7 states max, 2 mixes max]
// load stats file
RO 0.00 ''
Setting outlier threshold for clustering
RO->LS /home/tomek/Speech/HTS-demo_AMU_PL_ILO/stats/qst001/ver1/cmp.stats
and loading state occupation stats
Stats loaded for 96182 models
Mean Occupation Count = 6.644242
TR 0
Adjusting trace level
(
^
Error { expected
ERROR [+7230] EdError: item list parse error
FATAL ERROR - Terminating program /usr/local/HTS-2.1/bin/HHEd
Error in /usr/local/HTS-2.1/bin/HHEd -A -B -C /home/tomek/Speech/HTS-demo_AMU_PL_ILO/configs/trn.cnf -D -T 1 -p -i -H /home/tomek/Speech/HTS-demo_AMU_PL_ILO/models/qst001/ver1/cmp/clustered.mmf -m -a 1.0 -w /home/tomek/Speech/HTS-demo_AMU_PL_ILO/models/qst001/ver1/cmp/clustered.mmf /home/tomek/Speech/HTS-demo_AMU_PL_ILO/edfiles/qst001/ver1/cmp/cxc_mgc.hed /home/tomek/Speech/HTS-demo_AMU_PL_ILO/data/lists/full.list
I suspect it is either caused by an error in question file (e.g. missing "}") or even more likely by segment names used for schwa and glotal stop in my corpus which are respectively "@" and "?". Is that possible? If it is the second case do I have to start the whole process from the very beginning after adjusting the corpus or is it possible to save some data (it is very time consuming - took almost 6 days to get to this point)?
Another question is how far did I get in the whole process, is the tree-based context clustering at least half of the whole training part (speaking in terms of time - I have seen the procedures left)? This information would be very helpful.
Thank you in advance
--
Best regards,
Tomasz Kuczmarski
e-mail: faqster@xxxxxxxxx
skype : faqster