[hts-users:00030] Re: problems with HTS learning
- Subject: [hts-users:00030] Re: problems with HTS learning
- From: Łukasz Osowski <lo@xxxxxx>
- Date: Fri, 09 Apr 2004 21:00:15 +0200
- User-agent: Mozilla Thunderbird 0.5 (Windows/20040207)
Hello again!
Thank You for answers.
According to Your suggestion i've changed my polish voice to use pau
instead of # for silence phone, then created utts (voice works good,
utts looks ok.).
Then i've build HTS stucture once again. 1,2,3 phases of Trainning goes
well until to pau phoneme where error "No patch found" stops procedure.
(I've changed source of HInit to display warninng only, and not stop,
this the same error which i had using # instead of pau).
Do have idea what did I wrong?
best reagards
Lukasz Osowski
Here is log from teaching (beginning from last well teached phoneme):
/home/luke/mowa/festival/htk/bin.linux/HInit -A -C
/var/log/luke/mowa/HTS-demo/configs/default.conf -D -T 1 -m 1 -v 1.0e-04
-I /var/log/luke/mowa/HTS-demo_IVO-PL-JL/labels/mono_jl.mlf -M
/var/log/luke/mowa/HTS-demo/hmms/ivo_pl_ivona/jl/cmp/HInit -S
/var/log/luke/mowa/HTS-demo_IVO-PL-JL/lists/data_ivo_pl_ivona_jl.scp -l
p -o p /var/log/luke/mowa/HTS-demo/proto/s5_m1_c75_p3.prt
HTK Configuration Parameters[5]
Module/Tool Parameter Value
# NATURALWRITEORDER TRUE
# NATURALREADORDER TRUE
# DVFLOORSCALE 0.010000
# APPLYVFLOOR TRUE
# TREEMERGE FALSE
Initialising HMM /var/log/luke/mowa/HTS-demo/proto/s5_m1_c75_p3.prt . . .
States : 2 3 4 5 6 (width)
Mixes s1: 1 1 1 1 1 ( 75 )
( 75 )
Num Using: 0 0 0 0 0
Mixes s2: 2 2 2 2 2 ( 1 )
( MSD <= 1 )
Num Using: 0 0 0 0 0
Mixes s3: 2 2 2 2 2 ( 1 )
( MSD <= 1 )
Num Using: 0 0 0 0 0
Mixes s4: 2 2 2 2 2 ( 1 )
( MSD <= 1 )
Num Using: 0 0 0 0 0
Parm Kind: USER
Number of owners = 1
SegLab : p
maxIter : 20
epsilon : 0.000100
minSeg : 1
Updating : Means Variances MixWeights/DProbs TransProbs
- system is PLAIN
2609 Observation Sequences Loaded
Starting Estimation Process
Iteration 1: Average LogP = 628.43945
Iteration 2: Average LogP = 651.76727 Change = 23.32782
Iteration 3: Average LogP = 655.29395 Change = 3.52667
Iteration 4: Average LogP = 656.19019 Change = 0.89624
Iteration 5: Average LogP = 656.70245 Change = 0.51227
Iteration 6: Average LogP = 657.09125 Change = 0.38879
Iteration 7: Average LogP = 657.38782 Change = 0.29657
Iteration 8: Average LogP = 657.54889 Change = 0.16107
Iteration 9: Average LogP = 657.62836 Change = 0.07947
Iteration 10: Average LogP = 657.65588 Change = 0.02753
Iteration 11: Average LogP = 657.66565 Change = 0.00977
Iteration 12: Average LogP = 657.67047 Change = 0.00482
Iteration 13: Average LogP = 657.67059 Change = 0.00012
Iteration 14: Average LogP = 657.67236 Change = 0.00177
Iteration 15: Average LogP = 657.67255 Change = 0.00018
Iteration 16: Average LogP = 657.67322 Change = 0.00067
Iteration 17: Average LogP = 657.67310 Change = -0.00012
Iteration 18: Average LogP = 657.67407 Change = 0.00098
Iteration 19: Average LogP = 657.67407 Change = 0.00000
Estimation converged at iteration 20
Output written to directory
/var/log/luke/mowa/HTS-demo/hmms/ivo_pl_ivona/jl/cmp/HInit
HTK Configuration Parameters[5]
Module/Tool Parameter Value
NATURALWRITEORDER TRUE
NATURALREADORDER TRUE
# DVFLOORSCALE 0.010000
# APPLYVFLOOR TRUE
# TREEMERGE FALSE
==================================================
Realtime 0: 0.42
==================================================
/home/luke/mowa/festival/htk/bin.linux/HRest -A -C
/var/log/luke/mowa/HTS-demo/configs/default.conf -D -T 1 -m 1 -v 1.0e-04
-u tmvw -I /var/log/luke/mowa/HTS-demo_IVO-PL-JL/labels/mono_jl.mlf -M
/var/log/luke/mowa/HTS-demo/hmms/ivo_pl_ivona/jl/cmp/HRest -S
/var/log/luke/mowa/HTS-demo_IVO-PL-JL/lists/data_ivo_pl_ivona_jl.scp -l
p /var/log/luke/mowa/HTS-demo/hmms/ivo_pl_ivona/jl/cmp/HInit/p
HTK Configuration Parameters[5]
Module/Tool Parameter Value
# NATURALWRITEORDER TRUE
# NATURALREADORDER TRUE
# DVFLOORSCALE 0.010000
# APPLYVFLOOR TRUE
# TREEMERGE FALSE
Reestimating HMM
/var/log/luke/mowa/HTS-demo/hmms/ivo_pl_ivona/jl/cmp/HInit/p . . .
States : 2 3 4 5 6 (width)
Mixes s1: 1 1 1 1 1 ( 75 )
( 75 )
Num Using: 0 0 0 0 0
Mixes s2: 2 2 2 2 2 ( 1 )
( MSD <= 1 )
Num Using: 0 0 0 0 0
Mixes s3: 2 2 2 2 2 ( 1 )
( MSD <= 1 )
Num Using: 0 0 0 0 0
Mixes s4: 2 2 2 2 2 ( 1 )
( MSD <= 1 )
Num Using: 0 0 0 0 0
Parm Kind: USER
Number of owners = 1
SegLab : p
MaxIter : 20
Epsilon : 0.000100
Updating : Transitions Means Variances
- system is PLAIN
2609 Examples loaded, Max length = 67, Min length = 7
Ave LogProb at iter 1 = 657.90552 using 2609 examples
Ave LogProb at iter 2 = 657.90826 using 2609 examples change = 0.00273
Ave LogProb at iter 3 = 657.90820 using 2609 examples change = -0.00006
Estimation converged at iteration 3
HTK Configuration Parameters[5]
Module/Tool Parameter Value
NATURALWRITEORDER TRUE
NATURALREADORDER TRUE
# DVFLOORSCALE 0.010000
# APPLYVFLOOR TRUE
# TREEMERGE FALSE
==================================================
Realtime 0: 0.33
==================================================
=============== pau ================
/home/luke/mowa/festival/htk/bin.linux/HInit -A -C
/var/log/luke/mowa/HTS-demo/configs/default.conf -D -T 1 -m 1 -v 1.0e-04
-I /var/log/luke/mowa/HTS-demo_IVO-PL-JL/labels/mono_jl.mlf -M
/var/log/luke/mowa/HTS-demo/hmms/ivo_pl_ivona/jl/cmp/HInit -S
/var/log/luke/mowa/HTS-demo_IVO-PL-JL/lists/data_ivo_pl_ivona_jl.scp -l
pau -o pau /var/log/luke/mowa/HTS-demo/proto/s5_m1_c75_p3.prt
HTK Configuration Parameters[5]
Module/Tool Parameter Value
# NATURALWRITEORDER TRUE
# NATURALREADORDER TRUE
# DVFLOORSCALE 0.010000
# APPLYVFLOOR TRUE
# TREEMERGE FALSE
Initialising HMM /var/log/luke/mowa/HTS-demo/proto/s5_m1_c75_p3.prt . . .
States : 2 3 4 5 6 (width)
Mixes s1: 1 1 1 1 1 ( 75 )
( 75 )
Num Using: 0 0 0 0 0
Mixes s2: 2 2 2 2 2 ( 1 )
( MSD <= 1 )
Num Using: 0 0 0 0 0
Mixes s3: 2 2 2 2 2 ( 1 )
( MSD <= 1 )
Num Using: 0 0 0 0 0
Mixes s4: 2 2 2 2 2 ( 1 )
( MSD <= 1 )
Num Using: 0 0 0 0 0
Parm Kind: USER
Number of owners = 1
SegLab : pau
maxIter : 20
epsilon : 0.000100
minSeg : 1
Updating : Means Variances MixWeights/DProbs TransProbs
- system is PLAIN
5641 Observation Sequences Loaded
Starting Estimation Process
Warning: ViterbiAlign: No path found in 1'th segment
Warning: ViterbiAlign: No path found in 2'th segment
Warning: ViterbiAlign: No path found in 3'th segment
Warning: ViterbiAlign: No path found in 4'th segment
Warning: ViterbiAlign: No path found in 5'th segment
Warning: ViterbiAlign: No path found in 6'th segment
Warning: ViterbiAlign: No path found in 7'th segment
.....
- Follow-Ups
-
- [hts-users:00031] Re: problems with HTS learning, Heiga Zen / Byung-Ha Chun
- References
-
- [hts-users:00025] problems with HTS learning, Łukasz Osowski
- [hts-users:00026] Re: problems with HTS learning, Heiga Zen / Byung-Ha Chun
- [hts-users:00027] Re: problems with HTS learning, Łukasz Osowski
- [hts-users:00028] Re: problems with HTS learning, Heiga Zen / Byung-Ha Chun