[hts-users:03150] Re: Fatal error "can't shift that many" with HTS-demo_CMU-ARCTIC-ADAPT

[Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>]

Hi,

If you use HTS-2.2, please use SPTK-3.4.1 and hts_engine API-1.05.
The combination of HTS-2.2, SPTK-3.5 and hts_engine API-1.06 is not checked yet.

Could you change configure options as follows?

./configure \
TRAINSPKR='bdl clb jmk slt' \
ADAPTSPKR=rms \
F0_RANGES='bdl 40 210 clb 130 260 jmk 50 180 slt 110 280 rms 40 200'

Regards,
Keiichiro Oura


> ./configure --with-tcl-search-path=/home/machine/hts/ActiveTcl/bin
> --with-hts-search-path=/home/machine/hts/htk/HTKTools
> --with-fest-search-path=/usr/share/festival/examples
> --with-sptk-search-path=/home/machine/hts/SPTK-3.5/programs
> --with-hts-engine-search-path=/home/machine/hts/hts_engine_API-1.06/bin
> ADAPTSPKR=rms SPEAKER=rms TRAINSPKR='bdl clb jmk rms slt'




2012/1/14  <winkler@xxxxxxxxxxxxx>:
>
> Also a good catch, and it probably would be an issue for my install if it
> got a little farther, but it isn't the reason for the error I'm seeing. When
> I run it with the correct argument name as follows:
>
> ./configure --with-tcl-search-path=/home/machine/hts/ActiveTcl/bin
> --with-hts-search-path=/home/machine/hts/htk/HTKTools
> --with-fest-search-path=/usr/share/festival/examples
> --with-sptk-search-path=/home/machine/hts/SPTK-3.5/programs
> --with-hts-engine-search-path=/home/machine/hts/hts_engine_API-1.06/bin
> ADAPTSPKR=rms SPEAKER=rms TRAINSPKR='bdl clb jmk rms slt'
>
> I still get the error "shift: 35: can't shift that many". Should my
> configure line above be working, or does anyone know if there another error
> in it? Any idea what "shift: 35: can't shift that many" is about? Thanks!
>
> On Jan 13, 2012, at 12:05 PM, Piccand, Sebastien wrote:
>
> Ah yes you are right, you can specify the environment variables after the
> ./configure line.
> Could it just be a typo? I am using ADAPTSPKR instead of ADAPTSPK.
> (Configure won’t complain if you use a wrong name for the environment
> variables)
>
> Seb
>
> ----------------------------------------------------------------------------
> Sébastien Piccand, Researcher
> Service Research Group
> DOCOMO Communications Laboratories Europe GmbH
> Landsberger Strasse 312
> 80687 Munich, Germany
> Phone:  +49-89-56824-247
> Web:     www.docomolab-euro.com
> Facebook:  www.facebook.com/docomoeurolabs
> E-mail:  piccand@xxxxxxxxxxxxxxxxxx
> ----------------------------------------------------------------------------
> Managing Directors (Geschaeftsfuehrer):
> Dr. Masami Yabusaki, Kazuo Sugiyama, Moriya Kumasaki
> Amtsgericht Muenchen, HRB 132976
>
> From: winkler@xxxxxxxxxxxxx [mailto:winkler@xxxxxxxxxxxxx]
> Sent: Freitag, 13. Januar 2012 00:42
> To: hts-users@xxxxxxxxxxxxxxx
> Subject: [hts-users:03147] Re: Fatal error "can't shift that many" with
> HTS-demo_CMU-ARCTIC-ADAPT
>
>
> Hi,
>
> Thanks for your suggestion. In the file INSTALL that is in the root
> directory of the demo, it instructs that the environment variables that are
> listed in ./configure -help should be passed via ./configure as arguments,
> excerpted:
>
>
>    You can change various parameters such as speech analysis conditions and
> model training conditions
>    through ./configure arguments.  For example
>
>    % ./configure MGCORDER=34 GAMMA=0 FREQWARP=0.0              (34-th order
> cepstrum)
>    % ./configure MGCORDER=34 GAMMA=0 FREQWARP=0.55             (34-th order
> Mel-cepstrum)
>
>    % ./configure MGCORDER=12 GAMMA=1 FREQWARP=0.0  LNGAIN=0    (12-th order
> LSP,     linear gain)
>    % ./configure MGCORDER=12 GAMMA=1 FREQWARP=0.0  LNGAIN=1    (12-th order
> LSP,     log gain)
>    % ./configure MGCORDER=12 GAMMA=1 FREQWARP=0.55 LNGAIN=1    (12-th order
> Mel-LSP, log gain)
>    % ./configure MGCORDER=12 GAMMA=3 FREQWARP=0.55 LNGAIN=1    (12-th order
> MGC-LSP, log gain)
>
>    % ./configure NSTATE=7 NITER=10 WFLOOR=5   (# of HMM states=7, # of EM
> iterations=10, mix weight floo$
>
>
> When I try the same three approaches with the variable settings, but this
> time exporting the environment variables before running ./configure just as
> you suggested instead of putting them at the end of ./configure, I get
> exactly the same results (that is, when I train with all voices I get the
> "can't shift that many" error and when I train with a subset of the voices
> either with or without sending a new voice order via F0_RANGES, I get the
> "Order of speakers in ALLSPKR (slt) doesn't match to that in F0_RANGES
> (rms)!" error).
>
> I think the issue may be more related to the "can't shift that many" error
> which may be occurring with an otherwise-good ./configure line for a reason
> as-yet unknown, or maybe that I am not setting F0_RANGES correctly, but I
> think that the environment variables are being passed through to the scripts
> whether they are exported before running ./configure or when added to
> ./configure as arguments, or I wouldn't see the same respective results when
> I use different environment variables as arguments as when I export them.
>
> Could anyone share with me a known-working ./configure line for adapting to
> rms as a sanity check? Any advice on how to troubleshoot/solve "can't shift
> that many"?
>
> On Jan 12, 2012, at 12:57 PM, Piccand, Sebastien wrote:
>
>
> Hi,
>
> I think the problem comes from the fact that you are confusing arguments and
> variables.
> ADAPTSPKR, SPEAKER, TRAINSPKR are environment variables.
> You can either set them previous to the call to the ./configure command
> with:
>>> export ADAPTSPKR=rms
>>> export TRAINSPKR=’bdl clb jmk slt’
>>> …
>>> ./configure –allyouroptions
>
> Or if you want to put them in the same command line, you have to set them
> before calling ./configure, e.g:
>>> ADAPTSPKR=rms; TRAINSPKR=’bdl clb jmk slt’; …; ./configure –options
>
> Regards,
> Seb
> ----------------------------------------------------------------------------
> Sébastien Piccand, Researcher
> Service Research Group
> DOCOMO Communications Laboratories Europe GmbH
> Landsberger Strasse 312
> 80687 Munich, Germany
> Phone:  +49-89-56824-247
> Web:     www.docomolab-euro.com
> Facebook:  www.facebook.com/docomoeurolabs
> E-mail:  piccand@xxxxxxxxxxxxxxxxxx
> ----------------------------------------------------------------------------
> Managing Directors (Geschaeftsfuehrer):
> Dr. Masami Yabusaki, Kazuo Sugiyama, Moriya Kumasaki
> Amtsgericht Muenchen, HRB 132976
>
> From: winkler@xxxxxxxxxxxxx [mailto:winkler@xxxxxxxxxxxxx]
> Sent: Donnerstag, 12. Januar 2012 00:22
> To: hts-users@xxxxxxxxxxxxxxx
> Subject: [hts-users:03145] Fatal error "can't shift that many" with
> HTS-demo_CMU-ARCTIC-ADAPT
>
>
>
> Hello,
>
> When I run the HTS-demo_CMU-ARCTIC-ADAPT demo on Ubuntu 11.10 Desktop,
> 32-bit using the following ./configure arguments for adaptation with rms
> instead of slt:
>
> ./configure --with-tcl-search-path=/home/machine/hts/ActiveTcl/bin
> --with-hts-search-path=/home/machine/hts/htk/HTKTools
> --with-fest-search-path=/usr/share/festival/examples
> --with-sptk-search-path=/home/machine/hts/SPTK-3.5/programs
> --with-hts-engine-search-path=/home/machine/hts/hts_engine_API-1.06/bin
> ADAPTSPK=rms SPEAKER=rms TRAINSPKR='bdl clb jmk rms slt'
>
> After many hours of error-free work successfully extracting MGC coefficients
> from the raw audio it always eventually fails in the script phase
> "Extracting log f0 sequence from raw audio" with this error repeating for
> every raw file:
>
> F0 search range: bdl [40-210]
> Extracting f0 from raw/bdl/cmu_us_arctic_bdl_a0001.raw
> Unable to open mixer /dev/mixer
> [....]
> Extracting f0 from raw/slt/cmu_us_arctic_slt_b0539.raw
> Unable to open mixer /dev/mixer
>
> In the middle of this, there is one specific file that gets another specific
> error as follows:
>
> Extracting f0 from raw/jmk/cmu_us_arctic_jmk_a0501.raw
> x2x : error: input data is over the range of type 'short'!
>
> and when all of the raw files have gotten the 'Unable to open mixer
> /dev/mixer' error, there is this fatal error at the end:
>
> shift: 35: can't shift that many
> make[1]: *** [lf0] Error 2
> make[1]: Leaving directory `/home/machine/hts/voices/ADAPT/data'
> make: *** [data] Error 2
>
> running 'tail log -f' after this gives the following output:
>
> tail log -f
> Synthesizing a speech waveform from
> /home/machine/hts/HTS-demo_CMU-ARCTIC-ADAPT/data/labels/gen/slt/cmu_us_arctic_slt_alice31.lab
> using hts_engine...done.
> Synthesizing a speech waveform from
> /home/machine/hts/HTS-demo_CMU-ARCTIC-ADAPT/data/labels/gen/slt/cmu_us_arctic_slt_alice32.lab
> using hts_engine...done.
> Synthesizing a speech waveform from
> /home/machine/hts/HTS-demo_CMU-ARCTIC-ADAPT/data/labels/gen/slt/cmu_us_arctic_slt_alice33.lab
> using hts_engine...done.
> Synthesizing a speech waveform from
> /home/machine/hts/HTS-demo_CMU-ARCTIC-ADAPT/data/labels/gen/slt/cmu_us_arctic_slt_alice34.lab
> using hts_engine...done.
> Synthesizing a speech waveform from
> /home/machine/hts/HTS-demo_CMU-ARCTIC-ADAPT/data/labels/gen/slt/cmu_us_arctic_slt_alice35.lab
> using hts_engine...done.
> Synthesizing a speech waveform from
> /home/machine/hts/HTS-demo_CMU-ARCTIC-ADAPT/data/labels/gen/slt/cmu_us_arctic_slt_alice36.lab
> using hts_engine...done.
> Synthesizing a speech waveform from
> /home/machine/hts/HTS-demo_CMU-ARCTIC-ADAPT/data/labels/gen/slt/cmu_us_arctic_slt_alice37.lab
> using hts_engine...done.
> Synthesizing a speech waveform from
> /home/machine/hts/HTS-demo_CMU-ARCTIC-ADAPT/data/labels/gen/slt/cmu_us_arctic_slt_alice38.lab
> using hts_engine...done.
> Synthesizing a speech waveform from
> /home/machine/hts/HTS-demo_CMU-ARCTIC-ADAPT/data/labels/gen/slt/cmu_us_arctic_slt_alice39.lab
> using hts_engine...done.
> Synthesizing a speech waveform from
> /home/machine/hts/HTS-demo_CMU-ARCTIC-ADAPT/data/labels/gen/slt/cmu_us_arctic_slt_alice40.lab
> using hts_engine...done.
>
> A search for "can't shift that many" in the last year gives the following
> results, but I wasn't able to pin down the issue in this case other than the
> fact that it might be shell-related:
>
> http://www.google.com/search?q=%22can't+shift+that+many#q=%22can%27t+shift+that+many&hl=en&tbs=clue:1,qdr:y
>
> I've repeatedly re-read ./configure -help and I've been looking for a
> mistake in my configure arguments but any changes I've made have resulted in
> the same or different fatal errors. For instance, if I run with the
> following ./configure line instead (removing rms from TRAINSPKR, since the
> default arguments omit slt from TRAINSPKR):
>
> ./configure --with-tcl-search-path=/home/machine/hts/ActiveTcl/bin
> --with-hts-search-path=/home/machine/hts/htk/HTKTools
> --with-fest-search-path=/usr/share/festival/examples
> --with-sptk-search-path=/home/machine/hts/SPTK-3.5/programs
> --with-hts-engine-search-path=/home/machine/hts/hts_engine_API-1.06/bin
> ADAPTSPK=rms SPEAKER=rms TRAINSPKR='bdl clb jmk slt'
>
> I get this error at the same script phase:
>
> Order of speakers in ALLSPKR (slt) doesn't match to that in F0_RANGES (rms)!
>
> I haven't been able to find a definition of ALLSPKR in the configure help or
> elsewhere, but I see in the script that emits this error that the order
> should be "bdl clb jmk rms slt" which I guess doesn't correspond to using
> TRAINSPKR without rms. I decided to try to give an argument for F0_RANGES
> that puts the voices in what I hope is the correct order:
>
> ./configure --with-tcl-search-path=/home/machine/hts/ActiveTcl/bin
> --with-hts-search-path=/home/machine/hts/htk/HTKTools
> --with-fest-search-path=/usr/share/festival/examples
> --with-sptk-search-path=/home/machine/hts/SPTK-3.5/programs
> --with-hts-engine-search-path=/home/machine/hts/hts_engine_API-1.06/bin
> ADAPTSPK=rms SPEAKER=rms TRAINSPKR='bdl clb jmk slt' F0_RANGES='bdl 40 210
> clb 130 260 jmk 50 180 slt 110 280 rms 40 200'
>
> But this resulted in the identical "Order of speakers" error.
>
> So, I guess my first question is what would be a known-working set of
> ./configure arguments for doing what I'm trying to do (adapt the demo with
> rms instead of slt). My second question, only in the event that one of those
> known-working sets of arguments matches one of the three that I've tried, is
> where to begin troubleshooting the errors I'm getting when I'm using the
> right arguments. Thank you!
>
>
>