[hts-users:02515] Re: Is this a bug of hts-2.1.1 straight demo?

2010/6/4 Sébastien Le Maguer <Sebastien.Le_maguer@xxxxxxxx>

Hello,

I've runned the demo with the old data extraction process (adapted to the version of STRAIGHT) and I get the same error. Have you solved it with your patch ?

Sebastien

Le 02.06.2010 09:26, Sébastien Le Maguer a écrit :

Hi,

I'm going to run the demo with your patch and I will tell you tomorrow
if it works.

For the warning, I've runned a test with festival 1.96 and they
disappeared so it must be this.

Sébastien

Le mercredi 02 juin 2010 à 00:26 +0800, 康世胤 a écrit :

I think I find why this error happens.

you are right.
the error is caused by the wrong mgc data in straight demo 2.1.1.

and this is due to the WRONG VERSION of straight we use.

in the straight demo of 2.1& 2.1.1, the required straight version is
"STRAIGHTV40pcode".
and it outputs the parameter 'sp' whose vector size is 513.

but in the straight demo of 2.1.1beta, the required straight version
is "STRAIGHTV40_007c".
and the vector size of sp is 1025.

so different opt of mcep should be given:

$ diff Makefile.in.2.1.1 Makefile.in.2.1.1beta
142c142
< $(MGCEP) -a $(FREQWARP) -m $(MGCORDER) -l 1024 -j 0 -f 0.0 -q 3>
mgc/$${base}.mgc; \
---

$(MGCEP) -a $(FREQWARP) -m $(MGCORDER) -l 2048 -j 0 -f 0.0 -q 3>

mgc/$${base}.mgc; \

"STRAIGHTV40_007c" has a more strict licence
compared to the trial version "STRAIGHTV40pcode".
maybe that is why the developers decide to use "STRAIGHTV40pcode".

2010/5/31 Sébastien Le Maguer<Sebastien.Le_maguer@xxxxxxxx>
Ok, I don't think it is a compiler's fault because I'm using
HTS 2.1 on
the same computer and it doesn't fail.

I will run a test with the data extracted by the old procedure
and
compare to the new one. I will tell you the result tomorrow
when I will
get them

Sebastien

Le lundi 31 mai 2010 à 21:05 +0800, 康世胤 a écrit :

> no, I never met warning -2331 in my log
>
> 2010/5/31 Sébastien Le Maguer<Sebastien.Le_maguer@xxxxxxxx>
> When I'm looking the log files I've found some error
messages
> in the fullcontext reestimation step which look like
that:
>
>
> WARNING [-2331] UpdateModels: ih^ng-dh
> +eh=r@1_3/A:0_1_2/B:1-0-3@1-1&1-6#1-3$1-4!0-1;0-1|
0/C:1+1
> +2/D:content_3/E:pps+1@1+3&1+2#0
> +1/F:content_1/G:10_7/H:6=3@2=1|L-L%/I:0=0/J:16
+10-2[82]
> copied: only 0 egs
>
> After it propagates in every embedded reestimation
step. Have
> you go this ?
>
> Sebastien
>
> Le samedi 29 mai 2010 à 10:25 +0200, Sébastien Le
Maguer a
> écrit :
>
> > I'm on Ubuntu and I'm using gcc
> >
> > Le 29.05.2010 01:45, 康世胤 a écrit :
> > > No, i go back for 2.1 now.
> > > Which compiler do you use? Mine is icc. Maybe it
is the compiler's fault?
> > >
> > > 2010/5/28, Sébastien Le
Maguer<Sebastien.Le_maguer@xxxxxxxx>:
> > >
> > >> I try to debug it. For the moment, i've found
that in the StepBack the
> > >> beta calculus lead to log(0) and so no path is
found. Have you got some
> > >> news ?
> > >>
> > >> Sebastien
> > >>
> > >> Le jeudi 20 mai 2010 à 16:15 +0800, 康世胤 a
écrit :
> > >>
> > >>> yes
> > >>>
> > >>>
> > >>>
> > >>> 2010/5/20 Sébastien Le
Maguer<Sebastien.Le_maguer@xxxxxxxx>
> > >>> Hi,
> > >>>
> > >>> I've got this error too. When you try
to run HMGenS do you
> > >>> have a
> > >>> segmentation fault?
> > >>>
> > >>> Regards,
> > >>> Sebastien
> > >>>
> > >>>
> > >>> Le mardi 18 mai 2010 à 13:43 +0800,
康世胤 a écrit :
> > >>>
> > >>> > A error occurred when I try to
train the straight demo with
> > >>> the
> > >>> > default configurations in this
section
> > >>> >
> > >>> >
> > >>>
> > >>>
===================================================================================
> > >>> > Start generating speech parameter
sequences (stc) at Mon May
> > >>> 17
> > >>> > 17:29:59 CST 2010
> > >>> >
> > >>>
> > >>>
===================================================================================
> > >>> >
> > >>> >
> > >>> >
> > >>> > Here is the error messages:
> > >>> >
> > >>> > ......
> > >>> > Pruning-On[1500.0 100.0 5000.0]
> > >>> > Generating Label alice01.lab
> > >>> > Total number of frames = 3172
> > >>> > PdfStream[1]: 3172 frames
> > >>> > PdfStream[2]: 3172 frames
> > >>> > PdfStream[3]: 3172 frames
> > >>> > EM-based parameter generation
(Hidden: MIX, STATE)
> > >>> > WARNING [-7324] StepBack: Bad
data or over pruning
> > >>> >
in /usr/local/hts-2.1.1/bin/HMGenS
> > >>> > WARNING [-9999] ParamGen: failed
to compute output prob
> > >>> > in /usr/local/hts-2.1.1/bin/HMGenS
> > >>> > Iteration 1: Average LogP =
-3.152585e+06
> > >>> > Error
in /usr/local/hts-2.1.1/bin/HMGenS -A -B
> > >>> >
> > >>> -C
> >
>>> /home/saser/work/official_demos-2.1.1/HTS-demo_CMU-ARCTIC-SLT_STRAIGHT/configs/syn.cnf
> > >>> -D -T 1 -S
> >
>>> /home/saser/work/official_demos-2.1.1/HTS-demo_CMU-ARCTIC-SLT_STRAIGHT/data/scp/gen.scp
> > >>> -t 1500 100 5000 -c 2 -H
> >
>>> /home/saser/work/official_demos-2.1.1/HTS-demo_CMU-ARCTIC-SLT_STRAIGHT/models/qst001/ver1/cmp/stc_all.mmf
> > >>> -N
> >
>>> /home/saser/work/official_demos-2.1.1/HTS-demo_CMU-ARCTIC-SLT_STRAIGHT/models/qst001/ver1/dur/stc_all.mmf
> > >>> -M
> >
>>> /home/saser/work/official_demos-2.1.1/HTS-demo_CMU-ARCTIC-SLT_STRAIGHT/gen/qst001/ver1/stc/2
> >
>>> /home/saser/work/official_demos-2.1.1/HTS-demo_CMU-ARCTIC-SLT_STRAIGHT/models/qst001/ver1/cmp/tiedlist
> >
>>> /home/saser/work/official_demos-2.1.1/HTS-demo_CMU-ARCTIC-SLT_STRAIGHT/models/qst001/ver1/dur/tiedlist
> > >>> >
> > >>> >
> > >>> >
> > >>> > Before the error, I got tens of
thousands of warnings like
> > >>> this:
> > >>> >
> > >>> > Converged (norm=3.774747e+01,
change=1.620419e-04).
> > >>> > WARNING [-9999] OutProb: output
prob at frame 2 is 0.0
> > >>> > in /usr/local/hts-2.1.1/bin/HMGenS
> > >>> > WARNING [-9999] OutProb: output
prob at frame 3 is 0.0
> > >>> > in /usr/local/hts-2.1.1/bin/HMGenS
> > >>> > WARNING [-9999] OutProb: output
prob at frame 4 is 0.0
> > >>> > in /usr/local/hts-2.1.1/bin/HMGenS
> > >>> > WARNING [-9999] OutProb: output
prob at frame 5 is 0.0
> > >>> > in /usr/local/hts-2.1.1/bin/HMGenS
> > >>> > ......
> > >>> > WARNING [-9999] OutProb: output
prob at frame 3169 is 0.0
> > >>> > in /usr/local/hts-2.1.1/bin/HMGenS
> > >>> > WARNING [-9999] OutProb: output
prob at frame 3170 is 0.0
> > >>> > in /usr/local/hts-2.1.1/bin/HMGenS
> > >>> > WARNING [-9999] OutProb: output
prob at frame 3171 is 0.0
> > >>> > in /usr/local/hts-2.1.1/bin/HMGenS
> > >>> > Average LogP = -9.993695e+09
> > >>> > Generating Label alice02.lab
> > >>> > Total number of frames = 2915
> > >>> > PdfStream[1]: 2915 frames
> > >>> > PdfStream[2]: 2830 frames
> > >>> > PdfStream[3]: 2915 frames
> > >>> >
> > >>> >
> > >>> >
> > >>> > and strange parameter generation
log like this: (see the
> > >>> change of
> > >>> > iteration 6)
> > >>> >
> > >>> > Generating Label alice01.lab
> > >>> > Total number of frames = 3172
> > >>> > PdfStream[1]: 3172 frames
> > >>> > PdfStream[2]: 3172 frames
> > >>> > PdfStream[3]: 3172 frames
> > >>> > Parameter generation considering
global variance (GV)
> > >>> > Optimization=NEWTON step size
(init=1.000, inc=1.20,
> > >>> dec=0.50)
> > >>> > HMM weight=1.00, GV weight=1.00
> > >>> > Stream: 1
> > >>> > Feature: all
> > >>> > Iteration 1: GV Obj =
-8.703781e+00 (HMM:-8.703781e+00
> > >>> > GV:-4.669074e-27)
> > >>> > Iteration 2: GV Obj =
-4.485174e+01 (HMM:-6.515020e+00
> > >>> > GV:-3.833672e+01) Change =
-36.147964
> > >>> > Iteration 3: GV Obj =
-2.060439e+01 (HMM:-3.695115e+00
> > >>> > GV:-1.690928e+01) Change =
-11.900611
> > >>> > Iteration 4: GV Obj =
-1.077798e+01 (HMM:-5.220877e+00
> > >>> > GV:-5.557108e+00) Change =
-2.074204
> > >>> > Iteration 5: GV Obj =
-8.299071e+00 (HMM:-6.717686e+00
> > >>> > GV:-1.581385e+00) Change =
0.404710
> > >>> > Iteration 6: GV Obj =
-2.659332e+15 (HMM:-2.115852e+07
> > >>> > GV:-2.659332e+15) Change =
-2659332467537417.500000
> > >>> > Iteration 7: GV Obj =
-1.661987e+14 (HMM:-5.289479e+06
> > >>> > GV:-1.661987e+14) Change =
-166198677105619.750000
> > >>> > Iteration 8: GV Obj =
-1.038622e+13 (HMM:-1.322297e+06
> > >>> > GV:-1.038622e+13) Change =
-10386216934943.804688
> > >>> > Iteration 9: GV Obj =
-6.489885e+11 (HMM:-3.305405e+05
> > >>> > GV:-6.489881e+11) Change =
-648988480495.767700
> > >>> > Iteration 10: GV Obj =
-4.054301e+10 (HMM:-8.262077e+04
> > >>> > GV:-4.054293e+10) Change =
-40543012815.267204
> > >>> > Iteration 11: GV Obj =
-2.531591e+09 (HMM:-2.065053e+04
> > >>> > GV:-2.531570e+09) Change =
-2531590533.724836
> > >>> > Iteration 12: GV Obj =
-1.579305e+08 (HMM:-5.162820e+03
> > >>> > GV:-1.579253e+08) Change =
-157930472.685939
> > >>> > Iteration 13: GV Obj =
-9.833805e+06 (HMM:-1.293318e+03
> > >>> > GV:-9.832512e+06) Change =
-9833797.165811
> > >>> > Iteration 14: GV Obj =
-6.099850e+05 (HMM:-3.271551e+02
> > >>> > GV:-6.096578e+05) Change =
-609976.683197
> > >>> > Iteration 15: GV Obj =
-3.754546e+04 (HMM:-8.622074e+01
> > >>> > GV:-3.745924e+04) Change =
-37537.164351
> > >>> > Iteration 16: GV Obj =
-2.279440e+03 (HMM:-2.629029e+01
> > >>> > GV:-2.253150e+03) Change =
-2271.141377
> > >>> > Iteration 17: GV Obj =
-1.405188e+02 (HMM:-1.145926e+01
> > >>> > GV:-1.290596e+02) Change =
-132.219760
> > >>> > Iteration 18: GV Obj =
-1.528113e+01 (HMM:-7.827290e+00
> > >>> > GV:-7.453843e+00) Change =
-6.982061
> > >>> > Iteration 19: GV Obj =
-8.570871e+00 (HMM:-6.957192e+00
> > >>> > GV:-1.613679e+00) Change =
-0.271800
> > >>> > Iteration 20: GV Obj =
-8.300441e+00 (HMM:-6.758615e+00
> > >>> > GV:-1.541825e+00) Change =
-0.001369
> > >>> > Converged (norm=3.456561e+03,
change=1.369426e-03).
> > >>> >
> > >>> >
> > >>> >
> > >>> > Here is my parameters for the demo
configuration, tcl&
> > >>> matlab are in
> > >>> > PATH so I do not specify them.
> > >>> >
> > >>> > $ ./configure
> > >>>
--with-sptk-search-path=/usr/local/SPTK-3.3/bin
> > >>> >
--with-hts-search-path=/usr/local/hts-2.1.1/bin
> > >>> >
> > >>>
--with-hts-engine-search-path=/usr/local/hts_engine_API-1.03/bin
> > >>> >
--with-openfst-search-path=/usr/local/openfst-1.1/bin
> > >>> >
--with-straight-path=/usr/local/STRAIGHTV40_007c
> > >>> >
> > >>> >
> > >>> >
> > >>> > Here is my parameters for the
hts-2.1.1 configuration
> > >>> >
> > >>> > $ ./configure SMAX=11 PAT_LEN=8192
> > >>> --prefix=/usr/local/hts-2.1.1
> > >>> > CC=icc CXX=icpc
> > >>> >
> > >>> >
> > >>> >
> > >>> > Here is my icc version
> > >>> >
> > >>> > $ icc -V
> > >>> > Intel(R) C Intel(R) 64 Compiler
Professional for
> > >>> applications running
> > >>> > on Intel(R) 64, Version 11.1
Build 20091130 Package ID:
> > >>> > l_cproc_p_11.1.064
> > >>> > Copyright (C) 1985-2009 Intel
Corporation. All rights
> > >>> reserved.
> > >>> > FOR NON-COMMERCIAL USE ONLY
> > >>> >
> > >>> >
> > >>> >
> > >>> >
> > >>> > My questions are:
> > >>> > 1. Why this error occurred? Is
this error a bug?
> > >>> > 2. Is the Warings and iteration
changes normal? I did not
> > >>> find this
> > >>> > kind of messages in the same
section of the non-straight
> > >>> demo log
> > >>> > file.
> > >>> >
> > >>> >
> > >>> >
> > >>> >
> > >>> > the orignal log file is attached.
> > >>> >
> > >>> >
> > >>> >
> > >>> > --
> > >>> > 康世胤
> > >>> > Shiyin Kang
> > >>> >
> > >>>
> > >>>
> > >>>
> > >>>
> > >>>
> > >>>
> > >>>
> > >>
> > >>
> > >>
> > >>
> > >>
> > >
> >
> >
>
>
>
>
>
> --
> 康世胤
> Shiyin Kang
>
> kangshiyin@xxxxxxxxx
> Mobile: +86-13810352340
> Tel: +86-10-62775251
>
> FIT 3-522
> Lab of Human-Computer Speech Interaction,
> Department of Computer Science& Technology,
> Tsinghua University, Beijing, 100084, P.R.China

--
康世胤
Shiyin Kang

kangshiyin@xxxxxxxxx
Mobile: +86-13810352340
Tel: +86-10-62775251

FIT 3-522
Lab of Human-Computer Speech Interaction,
Department of Computer Science& Technology,
Tsinghua University, Beijing, 100084, P.R.China