[hts-users:00339] Re: about delta

["Nicholas Volk" <nvolk@xxxxxxxxxx>]

Hi,

As far as I know, the maximum process size on 32-bit Windows is
2MB. I've seen 1,6MB allocated to HEREst just before a crash.
This has caused me problems sometimes on Windows, while the same
training data worked well on Linux. However I was using
proprietary databases at the time.
I guess that this causes the problem.
64-bit Windows might work better (or does it: gcc is still a 32-bit
compiler).

When I last built Festival on Cygwin (in 2005), it went ok.
I am aware that latest 3.4.x and/or  latest version of Festival
might be problems under Cygwin. You might try an older
version of Festival. 1.4.4 should be fine unless the problems
are gcc or Festival compability related.

I'm assuming that you used Festival 1.9.5.
The unofficial 1.9.6 version of Festival might or might not also
work on Cygwin.

regards,
  Nicholas
> On Wed, 29 Mar 2006 15:51:25 +0200, Heiga ZEN (Byung Ha CHUN)
> <zen@xxxxxxxxxxxxxxxx> wrote:
>
>> Hi,
>>
>> Does anyone know where we should modify SPTK and HTS to run them on
>> Windows?
>> If you know, could you please show them on the mailing list?
>> We'd like to reflect these modifications to the next SPTK and HTS
>> release :-)
>>
>> Best regards,
>>
>> Heiga Zen (Byung Ha Chun)
>>
>
> Hi,
>
> I tried to build and build and run HTS on Windows using Cygwin.
>
> Configuration:
> AMD Athlon 64 3800+
> Microsoft Windows XP Professional Version 2002 Service Pack 2
> CYGWIN_NT-5.1 1.5.18(0.132/4/2) i686
> LC_ALL=C
> gcc (GCC) 3.4.4 (cygming special) (gdc 0.12, using dmd 0.125)
>
> Here are my experiences:
>
> 1. SPTK 3.0
> - Builds without modification. I removed xgr from build, because I have no
> X11 installed and its not used by HTS anyway.
>
> 2. HTS 2.0RC1
> - Applied the HTS-2.0RC1_for_HTK-3.3 patch to HTK source code (zip archive
> for Windows users). Two chunks failed (htk_htktools_nt.mkf and
> htk_htklib_nt.mkf) so I applied the patch manually to those files.
> - Followed the HTK on Windows installation instructions found at
> http://www.phon.ucl.ac.uk/resource/sfs/howto/htk.htm (setting environment,
> configure, patch of HTKLib/makefile and HTKTools/makefile, make)
>
> 3. Festival
> - Followed the Festival on windows installation instructions found at
> http://www.seas.upenn.edu/~mfickett/festival.html, this requires a
> Microsoft C++ compiler (VC98). I did not succeed to compile Festival using
> gcc (failed with  some STL related error messages).
>
> 4. HTS-demo_CMU-ARCTIC-SLT.RC1
> - Everything works fine until HERest crashes during embedded re-estimation
> (ERST1 in Training.pl). Here is the stack trace:
>
> Program received signal SIGSEGV, Segmentation fault.
> 0x610c4914 in memcpy () from /usr/bin/cygwin1.dll
> (gdb) where
> #0  0x610c4914 in memcpy () from /usr/bin/cygwin1.dll
> #1  0x6100275e in toascii () from /usr/bin/cygwin1.dll
> #2  0x61050c8e in lsearch () from /usr/bin/cygwin1.dll
> #3  0x610516e4 in lsearch () from /usr/bin/cygwin1.dll
> #4  0x61051f79 in lsearch () from /usr/bin/cygwin1.dll
> #5  0x610ab36c in setstate () from /usr/bin/cygwin1.dll
> #6  0x6104ed0a in lsearch () from /usr/bin/cygwin1.dll
> #7  0x610844ff in cygwin1!aclcheck () from /usr/bin/cygwin1.dll
> #8  0x0040545a in New (x=0x46c9c8, size=16) at HMem.c:420
> #9  0x00429cf7 in AttachWtTrAccs (hset=0x0, x=0x46c9c8) at HFB.c:232
> #10 0x0042a108 in InitialiseForBack (fbInfo=0x769390, x=0x46c9c8,
>      hset=0x22e890, uset=15, pruneInit=1.0000000200408773e+20, pruneInc=0,
>      pruneLim=1.0000000200408773e+20, minFrwdP=1.00000002e+20,
> calcDur=FALSE,
>      useAlign=FALSE) at HFB.c:338
> #11 0x00401a98 in Initialise () at HERest.c:708
> #12 0x00403df9 in main () at HERest.c:542
>
> Here I got stuck. Memory allocation in HTK causes a segmentation fault.
> Could be a bug in the HTK memory management that occures only on this
> platform/configuration. Or maybe the input data at this training stage is
> somehow erroneous. Did someone have a similar problem?
>
> A reference log file from a successful training session would be a great
> help for testing and debugging HTS on different systems. It would be easy
> to spot the location of the error by applying a diff between the reference
> and the recent log file. Can anybody provide such a reference log file?
>
>
> Best regards,
> Matthias
>