>  > Be aware that there's a big in v3_20 (and earlier) that causes
>  > problems if there are a lot of reject frames. Specifically if the
>  > fraction of frames rejected ever passes the fraction of work being
>  > done by a given thread (e.g. 12.5% reject with 8 threads). There's
>  > been a fix around for a while but I haven't had time to release it (or
>  > merge Arlo's recent improvements).
> Noted. Thank goodness I'm an accepting kind of guy, and don't tend
> to reject much. :-)

Rejecting frames is no joke! We've recently found that it's a great way
to speed up training if you reject them in such a way as to leave a
subset of data that has a uniform distribution over classes.  This can
reduce training time by an order of magnitude.

Unfortunately there's some issues with the multi-threading involved when
you're rejecting the majority of your data, due to the way in which
frames in a bunch are distributed to the threads.  The current best hack
is to dupe Quicknet into thinking that you have a lot more threads than
you're actually going to use.  For example, if you reject 90% of your
data and you want to train on a 4-CPU server, you should set
mlp_threads=40.  This is probably something that should be fixed in a
less crude manner... but it works for now!

-arlo

On Oct 24, 2007, at 9:17 PM, Arlo Faria wrote:

> Yes, Matlab-formatted weights are definitely the way to go.
> Specifically, you just do this in Matlab:
>
> >> load -mat qnstrn_exp/3layerMLP.weights
> >> clear bias3 weights23
> >> save -v4 qnstrn_exp/2layerMLP.weights
>
Wow!  That ain't rocket science!  Thanks for the tip.

> In addition to Barry's HATs setup, you can do neat things like
> "bottleneck" features from Brno: train a 5-layer MLP with a very
> narrow
> middle layer, and run your forward pass only up to that third layer.
> By the way, David mentioned that I have some new features that
> might one
> day be included in the next release. These allow online utterance
> padding and speaker-based mean/variance normalization, which are very
> handy in preparing various kinds of inputs to the nets.
Very cool, looking forward to it.

-Eric

On Oct 24, 2007, at 8:53 PM, David Johnson wrote:

>
> Eric> Or, does someone have some code to munge the weight files
> Eric> for me?
>
> I think it's a weight-file-munging exercise. Note that the new
> weightfile format is matlab v4-compatible, so the munging should be
> pretty straightforward. qncopywts will do this one day!!
aha.  I still have old rap3 weights (from qnstrn) so I can use
qncopywts to convert to matlab.  I guess I could hack the rap3
weights file rather easily (it's just lopping off the end, if I
remember) -- will qnmultifwd accept hacked rap3 weights for a 2-layer
network?  Or is rap3 meant to really only represent 3-layer networks?

> Eric> I have a wacky idea that I want to try out but it
> Eric> presupposes getting the output from the hidden layer. I can
> Eric> write the code to do this, but I figured that someone must
> Eric> have done this already (particularly, maybe Barry as part of
> Eric> the HATS experiments?).
>
> Eric> Also on my wish list: has anyone interfaced the feacat HTK
> Eric> reading code with quicknet? Looks like it would be an easy
> Eric> job to make a "htklist" input given the presence of
> Eric> "srilist" but I don't want to duplicate effort.
>
> I'm not aware of an htklist type. I had in the back of my mind that
> there were issues with "srilist" - maybe something to do with writing
> srilist files not working in feacat, or maybe something more
> relevant?? Indeed I can't remember doing anything with srilist files
> and qn* myself.
What I had meant was that you can convert via feacat between pfiles
and lists of htk files (or vice versa) using

feacat -ip pfile -op htk -olist output.list input.pf

feacat -ip htk -op pfile -o output.pf -lists input.list

(it isn't "htklist" per se as an argument, but using htk as an output
type implies lists of files).  I saw that srilist is an option to
qnmulti{trn,fwd} so I was hoping that this worked as advertised (i.e.
not having to convert to pfiles to train).  I think having the
ability to train directly from htk files would be extremely useful
for compatibility.

I'd be willing to take a crack at this but it's probably not going to
happen before January at the rate I'm going.

[BTW -- we have plans to port our Conditional Random Field
recognition software over to C++ using the qn libraries.  The current
java version understands pfiles, but it would be nice to have more
interoperability.  Again, not going to happen soon.]
> Be aware that there's a big in v3_20 (and earlier) that causes
> problems if there are a lot of reject frames. Specifically if the
> fraction of frames rejected ever passes the fraction of work being
> done by a given thread (e.g. 12.5% reject with 8 threads). There's
> been a fix around for a while but I haven't had time to release it (or
> merge Arlo's recent improvements).
Noted.  Thank goodness I'm an accepting kind of guy, and don't tend
to reject much. :-)

(Actually, come to think of it, we may have seen this bug.... hmmm....)

Thanks for the feedback,
-Eric

Yes, Matlab-formatted weights are definitely the way to go.
Specifically, you just do this in Matlab:

  >> load -mat qnstrn_exp/3layerMLP.weights
  >> clear bias3 weights23
  >> save -v4 qnstrn_exp/2layerMLP.weights

In addition to Barry's HATs setup, you can do neat things like
"bottleneck" features from Brno: train a 5-layer MLP with a very narrow
middle layer,  and run your forward pass only up to that third layer.

By the way, David mentioned that I have some new features that might one
day be included in the next release.  These allow online utterance
padding and speaker-based mean/variance normalization, which are very
handy in preparing various kinds of inputs to the nets.

-arlo

David Johnson wrote:
>
>
>  >>>>> "Eric" == Eric <fosler@...
> <mailto:fosler%40cse.ohio-state.edu>> writes:
>
> Eric> Hi all, I started looking at all of the new options (woo
> Eric> hoo) in v 3.20, and am curious: is there a way on the
> Eric> command line to take a 3-layer MLP and strip off the output
> Eric> layer (thus leaving the input-hidden layer) for forward
> Eric> processing purposes?
>
> Eric> Or, does someone have some code to munge the weight files
> Eric> for me?
>
> I think it's a weight-file-munging exercise. Note that the new
> weightfile format is matlab v4-compatible, so the munging should be
> pretty straightforward. qncopywts will do this one day!!
>
> Eric> I have a wacky idea that I want to try out but it
> Eric> presupposes getting the output from the hidden layer. I can
> Eric> write the code to do this, but I figured that someone must
> Eric> have done this already (particularly, maybe Barry as part of
> Eric> the HATS experiments?).
>
> Eric> Also on my wish list: has anyone interfaced the feacat HTK
> Eric> reading code with quicknet? Looks like it would be an easy
> Eric> job to make a "htklist" input given the presence of
> Eric> "srilist" but I don't want to duplicate effort.
>
> I'm not aware of an htklist type. I had in the back of my mind that
> there were issues with "srilist" - maybe something to do with writing
> srilist files not working in feacat, or maybe something more
> relevant?? Indeed I can't remember doing anything with srilist files
> and qn* myself.
>
> Be aware that there's a big in v3_20 (and earlier) that causes
> problems if there are a lot of reject frames. Specifically if the
> fraction of frames rejected ever passes the fraction of work being
> done by a given thread (e.g. 12.5% reject with 8 threads). There's
> been a fix around for a while but I haven't had time to release it (or
> merge Arlo's recent improvements).
>
> David.
>
> Eric> -Eric
>
> Eric> Yahoo! Groups Links
>
>

>>>>> "Eric" == Eric  <fosler@...> writes:

     Eric> Hi all, I started looking at all of the new options (woo
     Eric> hoo) in v 3.20, and am curious: is there a way on the
     Eric> command line to take a 3-layer MLP and strip off the output
     Eric> layer (thus leaving the input-hidden layer) for forward
     Eric> processing purposes?

     Eric> Or, does someone have some code to munge the weight files
     Eric> for me?

I think it's a weight-file-munging exercise.  Note that the new
weightfile format is matlab v4-compatible, so the munging should be
pretty straightforward.  qncopywts will do this one day!!

     Eric> I have a wacky idea that I want to try out but it
     Eric> presupposes getting the output from the hidden layer.  I can
     Eric> write the code to do this, but I figured that someone must
     Eric> have done this already (particularly, maybe Barry as part of
     Eric> the HATS experiments?).

     Eric> Also on my wish list: has anyone interfaced the feacat HTK
     Eric> reading code with quicknet?  Looks like it would be an easy
     Eric> job to make a "htklist" input given the presence of
     Eric> "srilist" but I don't want to duplicate effort.

I'm not aware of an htklist type.  I had in the back of my mind that
there were issues with "srilist" - maybe something to do with writing
srilist files not working in feacat, or maybe something more
relevant??  Indeed I can't remember doing anything with srilist files
and qn* myself.

Be aware that there's a big in v3_20 (and earlier) that causes
problems if there are a lot of reject frames.  Specifically if the
fraction of frames rejected ever passes the fraction of work being
done by a given thread (e.g. 12.5% reject with 8 threads).  There's
been a fix around for a while but I haven't had time to release it (or
merge Arlo's recent improvements).

David.

     Eric> -Eric





     Eric> Yahoo! Groups Links

Hi all,

I started looking at all of the new options (woo hoo) in v 3.20, and
am curious: is there a way on the command line to take a 3-layer MLP
and strip off the output layer (thus leaving the input-hidden layer)
for forward processing purposes?

Or, does someone have some code to munge the weight files for me?

I have a wacky idea that I want to try out but it presupposes getting
the output from the hidden layer.  I can write the code to do this,
but I figured that someone must have done this already (particularly,
maybe Barry as part of the HATS experiments?).

Also on my wish list: has anyone interfaced the feacat HTK reading
code with quicknet?   Looks like it would be an easy job to make a
"htklist" input given the presence of "srilist" but I don't want to
duplicate effort.

-Eric

I've released noway version 2.9.2, based on the source in ICSI's
"/u/drspeech/src/noway-2.9/" with the addition of some fixes from Dave
Gelbart to sort out compilation problems with more recent versions of
gcc.  I've also put the source in the drspeech CVS repository (with
this version tagged v2_9_2) and the corresponding tar file is
available at ftp://ftp.icsi.berkeley.edu/pub/real/davidj/noway.tar.gz.

The executable is installed for i586-linux and sun4-sunos5 (with the
sun4-sunos5 version built with gcc).  It is _not_ the default version
of noway yet, but rather is available as "noway-2.9.2" in the relevent
bin and man directories.

I've included the recent addition to the CHANGES file below.

David.

--
David Johnson, davidj@..., (510) 666-2983
System Manager, International Computer Science Institute
================================================================
============
noway-2.9.2:
============
Changes in 2.9.2 from 2.9.1 include:
    * Fixed Makefile to work with recent versions of gcc
      (David Gelbart).
    * Attempts to bring CHANGES file up to date
    * Fixed some broken-looking asserts in NW_acoustic.cc
============
noway-2.9.1:
============
No information is available on noway version 2.9.1. It was never released
but corresponds to the version of noway in /u/drspeech/src/noway-2.9/
as of August 2007 and is the first version checked into the ICSI drspeech
CVS repository.

Note that it's not called version 2.9 as there may already
be multiple differing versions around claiming to be version 2.9.

Differences for noway-2.7 include:
    * Addition of "-pause_phone_deletion_penalty"
    * Bugfixes???

============
noway-2.9:
============
No information is available on noway version 2.9.

============
noway-2.8:
============
No information is available on noway version 2.8.

============
noway-2.7:
============

================================================================

I've released feacalc v0_91.  It's installed in /u/drspeech/ for
sun4-sunos5 and i586-linux as feacalc-0_91 and wavs2onlaudio-0_91.  It
is _not_ the default version of feacalc but will be sometime soon.
The CVS tag is "v0_91".

The only signficant change is a patch by Dan Ellis - see
below. There's also an "INSTALL-icsi" file that details how to do a
release and install at ICSI.

====================
=== feacalc 0.91 ===
====================

Version 0.91 adds one feature:
  - Added the -numFilts argumen to control the number of filters in the
    auditory filter bank (correspons to the rasta -c option)
  - Added the INSTALL-icsi file

The .tar.gz is at
ftp://ftp.icsi.berkeley.edu/pub/real/davidj/feacalc.tar.gz.

David.

>>>>> "Scott" == Scott Otterson <scotto@...> writes:

     Scott> <!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
     Scott> <html>
     Scott> <head>
     Scott> </head>
     Scott> <body bgcolor="#ffffff" text="#000000">


     Scott> Hi,
     Scott>
     Scott> There's a quicknet 3_20 that shows up when you try to
            compile with gcc

This has already been fixed in the source, although I haven't released
a new version with it in yet.  I'll include the patch below.

David.

Index: QN_fir.h
===================================================================
RCS file: /u/drspeech/repos/quicknet2/QN_fir.h,v
retrieving revision 1.4
retrieving revision 1.5
diff -u -r1.4 -r1.5
--- QN_fir.h 21 Feb 2007 23:32:22 -0000 1.4
+++ QN_fir.h 13 Jun 2007 18:34:44 -0000 1.5
@@ -1,5 +1,5 @@
  // -*- c++ -*-
-// $Header: /u/drspeech/repos/quicknet2/QN_fir.h,v 1.4 2007/02/21 23:32:22
davidj Exp $
+// $Header: /u/drspeech/repos/quicknet2/QN_fir.h,v 1.5 2007/06/13 18:34:44
davidj Exp $

  #ifndef QN_fir_h_INCLUDED
  #define QN_fir_h_INCLUDED
@@ -113,9 +113,9 @@

      // Class statics to generate standard delta filter kernels
      // build a filter for delta calculation - write <len> pts into fv[]
-    static int QN_InFtrStream_FIR::FillDeltaFilt(float *fv, size_t len);
+    static int FillDeltaFilt(float *fv, size_t len);
      // build a filter for double-delta calculation - convol'n of 2 dfilts
-    static int QN_InFtrStream_FIR::FillDoubleDeltaFilt(float *fv, size_t len);
+    static int FillDoubleDeltaFilt(float *fv, size_t len);

  private:
      QN_ClassLogger clog;  // Logging object.

Scott> 4.21.  The error message is:
     Scott> <blockquote><tt>In file included from QN_utils.cc:53:
     Scott> QN_fir.h:116: error: extra qualification 'QN_InFtrStream_FIR::' on
     Scott> member 'FillDeltaFilt'
     Scott> QN_fir.h:118: error: extra qualification 'QN_InFtrStream_FIR::' on
     Scott> member 'FillDoubleDeltaFilt'
     Scott> QN_utils.cc: In function 'void QN_readwrite_weights(int, const char*,
     Scott> QN_FileMode, QN_MLPWeightFile&amp;, QN_MLP&amp;, float*, float*)':
     Scott> QN_utils.cc:468: warning: deprecated conversion from string constant
to
     Scott> 'char*'
     Scott> QN_utils.cc:470: warning: deprecated conversion from string constant
to
     Scott> 'char*'
     Scott> make: *** [QN_utils.o] Error 1</tt>
     Scott> </blockquote>
     Scott> which can be fixed by the following changes to QN_fir.h:
     Scott> <blockquote><tt>//&nbsp;&nbsp;&nbsp; static int
     Scott> QN_InFtrStream_FIR::FillDeltaFilt(float *fv, size_t len);
     Scott> &nbsp;&nbsp;&nbsp; static int FillDeltaFilt(float *fv, size_t len);
     Scott> &nbsp;&nbsp;&nbsp; // build a filter for double-delta calculation -
convol'n of 2
     Scott> dfilts
     Scott> //&nbsp;&nbsp;&nbsp; static int
QN_InFtrStream_FIR::FillDoubleDeltaFilt(float *fv,
     Scott> size_t len);
     Scott> &nbsp;&nbsp;&nbsp; static int FillDoubleDeltaFilt(float *fv, size_t
len);
     Scott>   </tt></blockquote>
     Scott> The problem is that a function's class name occurs within the class
     Scott> declaration.
     Scott>
     Scott> Scott
     Scott>
     Scott>
gnnp

In file included from QN_utils.cc:53:
QN_fir.h:116: error: extra qualification 'QN_InFtrStream_FIR::' on member 'FillDeltaFilt'
QN_fir.h:118: error: extra qualification 'QN_InFtrStream_FIR::' on member 'FillDoubleDeltaFilt'
QN_utils.cc: In function 'void QN_readwrite_weights(int, const char*, QN_FileMode, QN_MLPWeightFile&, QN_MLP&, float*, float*)':
QN_utils.cc:468: warning: deprecated conversion from string constant to 'char*'
QN_utils.cc:470: warning: deprecated conversion from string constant to 'char*'
make: *** [QN_utils.o] Error 1

//    static int QN_InFtrStream_FIR::FillDeltaFilt(float *fv, size_t len);
    static int FillDeltaFilt(float *fv, size_t len);
    // build a filter for double-delta calculation - convol'n of 2 dfilts
//    static int QN_InFtrStream_FIR::FillDoubleDeltaFilt(float *fv, size_t len);
    static int FillDoubleDeltaFilt(float *fv, size_t len);

>>>>> "Petr" == Petr Fousek <p.fousek@...> writes:

     Petr> Hi, I found a small bug in cmdline processing in labcat. One
     Petr> cannot abbreviate "-lists" option to "-l".

     Petr> Example: - converting per-file pfiles containing targets to
     Petr> one big pfile:

     Petr> works: labcat -ip pfile -op pfile -lists -i list.txt -o
     Petr> tgts.pfile

     Petr> does NOT work: labcat -ip pfile -op pfile -l -i list.txt -o
     Petr> tgts.pfile

This is because labcat has the "-l?abrange" option, which clashes with
"-l?ist".  I think the fix is to require make them be "-li?st" and
"-la?brange" for both labcat and feacat.  And maybe leave "-l?ist" as
an option for feacat, to minimize broken scripts, but only document
the "-li?st" option in the man pages.

Does this sound sensible?

David.

     Petr> Note that feacat is OK.

     Petr> Petr.




     Petr> Yahoo! Groups Links

Hi,

I found a small bug in cmdline processing in labcat. One cannot
abbreviate "-lists" option to "-l".

Example:
- converting per-file pfiles containing targets to one big pfile:

works:
labcat -ip pfile -op pfile -lists -i list.txt -o tgts.pfile

does NOT work:
labcat -ip pfile -op pfile -l -i list.txt -o tgts.pfile

Note that feacat is OK.

Petr.

On Jun 18, 2007, at 5:17 PM, David Johnson wrote:

> Not so sure about the dpwetcl stuff, especially as I don't know much
> about what labels2pfile adds to the feacat/labcat/pfile_utils mix.
I believe that it had the ability to convert from label files (like
those of the TIMIT/XWAVES files) into pfile labels.  However, the
code is (a) really old, and (b) can be pretty easily re-written as a
standalone script to interact with the labcat script that we don't
use this anymore at OSU.

The main thing to keep in mind is that you need to match the
windowing of feacalc (or whatever you're using to produce features).
So, if you use 25 msec windows with a 10 msec step, then you'll need
to take whatever labels are at 12.5 msec, 22.5 msec, 32.5 msec, etc.
An option is to have the script output ascii labels per frame, and
then feed directly into labcat with the -deslen file as your pfile
and -dltype pfile.

I suppose that it would be a good idea to make a version of labcat
(or labcalc?) that converts MLF/TIMIT/XWAVES label files to  for the
novice; we could put it on the wish list and hope that someone takes
it on.  But doing any sort of string processing of (possibly ill-
formatted files) in C++ makes me break out in hives.  Pass the
calamine lotion!

-Eric

>>>>> "yaoxuchen" == yaoxuchen  <yaoxuchen@...> writes:

     yaoxuchen> I'd strongly suggest not using SPRACHCore unless
     yaoxuchen> absolutely necessary as it's been many, many years
     yaoxuchen> since it's had any maintenance. Most of the important
     yaoxuchen> stuff is maintained separately now. Dan Ellis did most
     yaoxuchen> of the work on SPRACHCore and he left ICSI many years
     yaoxuchen> ago.

     yaoxuchen> What do folks still need from SPRACHcore that can't be
     yaoxuchen> obtained in other ways?

     yaoxuchen>  [Xuchen] I checked my QuickNet installation, and found
     yaoxuchen> that I use the following packages in SPRACHcore:

     yaoxuchen> 1. SPRACHcore-2004-08-26/guitools/dpwetcl package, it's
     yaoxuchen> used by labels2pfile.  2. SPRACHcore-2004-08-26/rasta
     yaoxuchen> package. In fact there is a rasta.tar.Z package in ICSI
     yaoxuchen> ftp, but it seems too old (1996), so I just use rasta
     yaoxuchen> from SPRACHcore.

There's definitely a newer version of rasta around, e.g. in
ftp://ftp.icsi.berkeley.edu/pub/real/dpwe/.

Not so sure about the dpwetcl stuff, especially as I don't know much
about what labels2pfile adds to the feacat/labcat/pfile_utils mix.

David.

     yaoxuchen> Xuchen





     yaoxuchen> Yahoo! Groups Links

>>>>> "Petr" == Petr Fousek <p.fousek@...> writes:

     Petr> Hi, I compiled pfile_utils v0.5 with QuickNet 3.20
     Petr> libraries. On some machines (probably 64bits), strange
     Petr> things happen. I am not a developper so I am asking rather
     Petr> you - QN gurus. PLS, don't miss the note at the end.

It looks like you're having problems because you're building on a 64
bit machine.  I'm not sure anyone has tested anything significantly
with 64 bit builds. I certainly haven't!  I've done some trivial tests
on QuickNet but more to make sure that it works as a library as
there's no strong need for the extra memory size with
feacat/feacalc/pfile_utils/qnstrain etc. but there may be for other
apps that use it.  Note that we do run code on plenty of 64 bit
machines at ICSI, but we build on 32 bit machines so we end up with 32
bit executables.  I'd suggest either doing that or using "-m 32"
during the build, which I'm sure will fix your problems.

That being said, the qnsfwd reporting is a 64-bit-specific bug that
I'll fix. I'll look into 64 bit issues with pfile_utils next time I
work on that package.

David.

     Petr> 1) pfile_create

     Petr> when I make a small ascii file "0 0 1.0<CR>0 1 2.0<CR>" and
     Petr> run pfile_create -f 1 -l 0 -i <file> -o <file.pfile>, it
     Petr> says: ERROR: Sentence numbers must be in sequence. Current
     Petr> sent = 0, next sent found = -1 With executables compiled on
     Petr> certain machines this command works as it is supposed to.

     Petr> machines working OK (i686): - SMP, XEON fam 15/mod 2,
     Petr> 2.40GHz, kernel 2.6.21.1, gcc version 3.4.4 20050721 (Red
     Petr> Hat 3.4.4-2) - Acer laptop, Pent-M, kernel 2.6.20, gcc
     Petr> 4.1.1. 20070105 (RH 4.1.1-51)

     Petr> problems on (x86_64): - SMP, XEON 5160, fam 6/mod 15,
     Petr> 3.00GHz, kernel 2.6.20-1.2307b, gcc version 4.1.1 20070105
     Petr> (Red Hat 4.1.1-51) - SMP, Pent4 fam 15/mod 4, 3.2GHz, kernel
     Petr> 2.6.18-1.2200.fc5, 4.1.1 20060525 (Red Hat 4.1.1-1)


     Petr> 2) quicknet

     Petr> - probably related. Tools qnstrn/qnsfwd from QN3.20, when
     Petr> they are running, they report no support for large files,
     Petr> although they DO support large files (tested with a 39GB
     Petr> pfile):

     Petr> Program version: qnsfwd v3_20.  Large PFiles: disabled.
     Petr> <------- Posix threads: enabled.  BLAS: enabled.  MLP type:
     Petr> QN_MLP_ThreadFl3.


     Petr> NOTE: I compiled the quicknet with the patches I posted in
     Petr> the thread
     Petr> http://tech.groups.yahoo.com/group/icsi-speech-tools/message/114
     Petr> without which I cannot compile QN3.20 at all.

     Petr> Any suggestions?

     Petr> Petr.




     Petr> Yahoo! Groups Links

Hi,

I compiled pfile_utils v0.5 with QuickNet 3.20 libraries. On some
machines (probably 64bits), strange things happen. I am not a
developper so I am asking rather you - QN gurus. PLS, don't miss the
note at the end.

1) pfile_create

when I make a small ascii file "0 0 1.0<CR>0 1 2.0<CR>" and run
pfile_create -f 1 -l 0 -i <file> -o <file.pfile>, it says:
ERROR: Sentence numbers must be in sequence. Current sent = 0, next
sent found = -1
With executables compiled on certain machines this command works as it
is supposed to.

machines working OK (i686):
- SMP, XEON fam 15/mod 2, 2.40GHz, kernel 2.6.21.1, gcc version 3.4.4
20050721 (Red Hat 3.4.4-2)
- Acer laptop, Pent-M, kernel 2.6.20, gcc 4.1.1. 20070105 (RH 4.1.1-51)

problems on (x86_64):
- SMP, XEON 5160, fam 6/mod 15, 3.00GHz, kernel 2.6.20-1.2307b, gcc
version 4.1.1 20070105 (Red Hat 4.1.1-51)
- SMP, Pent4 fam 15/mod 4, 3.2GHz, kernel 2.6.18-1.2200.fc5, 4.1.1
20060525 (Red Hat 4.1.1-1)


2) quicknet

- probably related. Tools qnstrn/qnsfwd from QN3.20, when they are
running, they report no support for large files, although they DO
support large files (tested with a 39GB pfile):

Program version: qnsfwd v3_20.
Large PFiles: disabled.      <-------
Posix threads: enabled.
BLAS: enabled.
MLP type: QN_MLP_ThreadFl3.


NOTE:
I compiled the quicknet with the patches I posted in the thread
http://tech.groups.yahoo.com/group/icsi-speech-tools/message/114
without which I cannot compile QN3.20 at all.

Any suggestions?

Petr.

I'd strongly suggest not using SPRACHCore unless absolutely necessary
as it's been many, many years since it's had any maintenance. Most of
the important stuff is maintained separately now. Dan Ellis did most
of the work on SPRACHCore and he left ICSI many years ago.

What do folks still need from SPRACHcore that can't be obtained in
other ways?

  [Xuchen] I checked my QuickNet installation, and found that I use the
following packages in SPRACHcore:

1. SPRACHcore-2004-08-26/guitools/dpwetcl package, it's used by
labels2pfile.
2. SPRACHcore-2004-08-26/rasta package. In fact there is a rasta.tar.Z
package in ICSI ftp, but it seems too old (1996), so I just use rasta from
SPRACHcore.

Xuchen

>>>>> "Petr" == Petr Fousek <p.fousek@...> writes:

     Petr> Olivier Bornet agreed to publish the patches, so here you
     Petr> are: http://eniac.hostuju.cz/patches

I've merged the QuickNet patches.

Specifically they:
  - fix a  declaration error in QN_FIR.h that broke some versions of
    gcc
  - fixed so atlas_buildinfo.h is only used if it's actually there as
    apparently that's not the case with all atlas installations.

They'll be in the next release.

David.

     Petr> Petr.




     Petr> --- In icsi-speech-tools@yahoogroups.com, "yaoxuchen"
     Petr> <yaoxuchen@...> wrote:
     >> You are welcome.
     >>
     >> Could you please give me a link about all the patches for gcc
     >> 4.1? I
     Petr> never
     >> know there's such kind of patches.
     >>
     >> Thanks.
     >>
     >> Xuchen
     >>
     >> ________________________________________ From:
     >> icsi-speech-tools@yahoogroups.com
     >> [mailto:icsi-speech-tools@yahoogroups.com] On Behalf Of Petr
     >> Fousek Sent: Wednesday, June 13, 2007 6:23 PM To:
     >> icsi-speech-tools@yahoogroups.com Subject: [icsi-speech-tools]
     >> Re: feacat 1.01 released ---- dpwelib dependence
     >>
     >> Xuchen, your configure trick succeded; you were right also with
     >> gcc version.
     >>
     >> Meanwhile I found a patch for dpwelib-2002-10-23 (again from
     >> Olivier Bornet) and the package compiled OK even with gcc
     >> 4.1.1. In summary, after patching dpwelib-2002-10-23 and
     >> quicknet-v3.20, all packages (QN3.20, feacat_1.01 and
     >> pfile_utils-v0.50) compiled OK with gcc 4.1.1.
     >>
     >> Anyway, thanks for your suggestions.
     >>
     >> Petr.
     >>
     >> --- In icsi-speech-tools@yahoogroups.com, "yaoxuchen"
     >> <yaoxuchen@> wrote:
     >> >
     >> > The configure script error you mentioned is probably caused
     >> by Tcl/Tk, if > your linux box is quite new.
     >> >
     >> > Use the following script before configure:
     >> >
     >> > cp configure{,.orig} && sed "s/relid'/relid/" configure.orig
     >> >
     Petr> configure
     >> > > You may also need a lower gcc version (i.e. 3.4 or 3.2) to
     >> compile dpwelib.
     >> >
     >> > Xuchen
     >> >
     >> > ________________________________________ > From:
     >> icsi-speech-tools@yahoogroups.com >
     >> [mailto:icsi-speech-tools@yahoogroups.com] On Behalf Of Petr
     >> Fousek > Sent: Friday, June 08, 2007 10:29 PM > To:
     >> icsi-speech-tools@yahoogroups.com > Subject:
     >> [icsi-speech-tools] Re: feacat 1.01 released ---- dpwelib >
     >> dependence
     >> >
     >> > SHORT: Where one can obtain dpwelib sources?
     >> >
     >> > LONGER: > - feacat 1.01 requires dpwelib and QuickNet3 libs
     >> to compile
     >> >
     >> > - QN version 3.20 compiled OK (thanks!), though some modified
     >> patches > from Olivier Bornet for QN v3.11 were needed
     >> >
     >> > - the most recent dpwelib I could find is
     >> dpwelib-2006-04-19. However, > this does neither compile on my
     >> current x64_64 linux box (configure > script syntax error) nor
     >> on an old Celeron. Where can one get dpwelib > which compiles
     >> so that feacat builds?
     >> >
     >> > Petr.
     >> >
     >> > --- In icsi-speech-tools@yahoogroups.com, David Johnson
     >> <davidj@> > wrote:
     >> > >
     >> > > Version 1.01 of the feacat package has been released,
     >> tagged in the > > CVS archive and installed in /u/drspeech/ at
     >> ICSI. feacat includes > > several versatile utilities for
     >> munging feature files, probability > > files, label files and
     >> line-based text files.
     >> > >
     >> > > This primary change in this releases is that disabling of
     >> "//" for > > specifying parallel feature files by default, and
     >> the associated > > "-usedoubleslash" option for enabling
     >> it. This brings it in line
     Petr> with
     >> > > the version of feacat used at SRI.
     >> > >
     >> > > From the NEWS file:
     >> > >
     >> > > ----------------------------------------------------------
     >> > > ==================== > > === feacat 1.01 ==== > >
     >> ====================
     >> > >
     >> > > - Added the "-usedoubleslash" option so the "//" syntax for
     >> parallel > > feature files isn't enabled by default (it causes
     >> more trouble than > > it's worth!) > > - Fixed "ascii" format
     >> so it uses %g for output and added "asciif" > which > > uses
     >> the version 1.00 and before %f output.  > > - Added config.sub
     >> and config.guess to dist.
     >> > > ----------------------------------------------------------
     >> > >
     >> > > The source distributions is available at:
     >> > >
     >> > >
     >> ftp://ftp.ICSI.Berkeley.EDU/pub/real/davidj/feacat-1.01.tar.gz
     >> > >
     >> > > The following programs have been installed in >
     >> /u/drspeech/i586-linux/bin/ > > and
     >> /u/drspeech/sun4-sunos5/bin/ at ICSI:
     >> > >
     >> > > feacat-1_01 > > labcat-1_01 > > linecat-1_01 > >
     >> segpostsum-1_01
     >> > >
     >> > > These versions were linked with QuickNet v3_20.
     >> > >
     >> > > The default version of feacat has not been changed but
     >> these new > > versions will become the default in the near
     >> future if there are no > > problems.
     >> > >
     >> > > Cheers,
     >> > >
     >> > > David.
     >> > >
     >> > > --
     >> > > David Johnson, davidj@, (510) 666-2983 > > International
     >> Computer Science Institute
     >> > >
     >> >
     >>




     Petr> Yahoo! Groups Links

>>>>> "Petr" == Petr Fousek <p.fousek@...> writes:

     >> Looking in the NEWS file, I don't see any mention of testing
     >> with >=2GB pfiles.  If it doesn't work, I'm guessing it just
     >> needs autoconf/header tweaking as all of the pfile writing is
     >> handled by the QuickNet routines.


     Petr> We had that problem too at IDIAP. Guys from Brno did the
     Petr> neccessary "tweaking". Check this page, search for "Note for
     Petr> SprachCore".

     Petr> http://www.fit.vutbr.cz/research/groups/speech/sw/trapper.html.cs

I'd strongly suggest not using SPRACHCore unless absolutely necessary
as it's been many, many years since it's had any maintenance.  Most of
the important stuff is maintained separately now.  Dan Ellis did most
of the work on SPRACHCore and he left ICSI many years ago.

What do folks still need from SPRACHcore that can't be obtained in
other ways?

David.

     Petr> Petr.





     Petr> Yahoo! Groups Links

Olivier Bornet agreed to publish the patches, so here you are:

http://eniac.hostuju.cz/patches

Petr.




--- In icsi-speech-tools@yahoogroups.com, "yaoxuchen" <yaoxuchen@...>
wrote:
>
> You are welcome.
>
> Could you please give me a link about all the patches for gcc 4.1? I
never
> know there's such kind of patches.
>
> Thanks.
>
> Xuchen
>
> ________________________________________
> From: icsi-speech-tools@yahoogroups.com
> [mailto:icsi-speech-tools@yahoogroups.com] On Behalf Of Petr Fousek
> Sent: Wednesday, June 13, 2007 6:23 PM
> To: icsi-speech-tools@yahoogroups.com
> Subject: [icsi-speech-tools] Re: feacat 1.01 released ---- dpwelib
> dependence
>
> Xuchen, your configure trick succeded; you were right also with gcc
> version.
>
> Meanwhile I found a patch for dpwelib-2002-10-23 (again from Olivier
> Bornet) and the package compiled OK even with gcc 4.1.1. In summary,
> after patching dpwelib-2002-10-23 and quicknet-v3.20, all packages
> (QN3.20, feacat_1.01 and pfile_utils-v0.50) compiled OK with gcc 4.1.1.
>
> Anyway, thanks for your suggestions.
>
> Petr.
>
> --- In icsi-speech-tools@yahoogroups.com, "yaoxuchen" <yaoxuchen@>
> wrote:
> >
> > The configure script error you mentioned is probably caused by
> Tcl/Tk, if
> > your linux box is quite new.
> >
> > Use the following script before configure:
> >
> > cp configure{,.orig} && sed "s/relid'/relid/" configure.orig >
configure
> >
> > You may also need a lower gcc version (i.e. 3.4 or 3.2) to compile
> dpwelib.
> >
> > Xuchen
> >
> > ________________________________________
> > From: icsi-speech-tools@yahoogroups.com
> > [mailto:icsi-speech-tools@yahoogroups.com] On Behalf Of Petr Fousek
> > Sent: Friday, June 08, 2007 10:29 PM
> > To: icsi-speech-tools@yahoogroups.com
> > Subject: [icsi-speech-tools] Re: feacat 1.01 released ---- dpwelib
> > dependence
> >
> > SHORT: Where one can obtain dpwelib sources?
> >
> > LONGER:
> > - feacat 1.01 requires dpwelib and QuickNet3 libs to compile
> >
> > - QN version 3.20 compiled OK (thanks!), though some modified patches
> > from Olivier Bornet for QN v3.11 were needed
> >
> > - the most recent dpwelib I could find is dpwelib-2006-04-19. However,
> > this does neither compile on my current x64_64 linux box (configure
> > script syntax error) nor on an old Celeron. Where can one get dpwelib
> > which compiles so that feacat builds?
> >
> > Petr.
> >
> > --- In icsi-speech-tools@yahoogroups.com, David Johnson <davidj@>
> > wrote:
> > >
> > > Version 1.01 of the feacat package has been released, tagged in the
> > > CVS archive and installed in /u/drspeech/ at ICSI. feacat includes
> > > several versatile utilities for munging feature files, probability
> > > files, label files and line-based text files.
> > >
> > > This primary change in this releases is that disabling of "//" for
> > > specifying parallel feature files by default, and the associated
> > > "-usedoubleslash" option for enabling it. This brings it in line
with
> > > the version of feacat used at SRI.
> > >
> > > From the NEWS file:
> > >
> > > ----------------------------------------------------------
> > > ====================
> > > === feacat 1.01 ====
> > > ====================
> > >
> > > - Added the "-usedoubleslash" option so the "//" syntax for parallel
> > > feature files isn't enabled by default (it causes more trouble than
> > > it's worth!)
> > > - Fixed "ascii" format so it uses %g for output and added "asciif"
> > which
> > > uses the version 1.00 and before %f output.
> > > - Added config.sub and config.guess to dist.
> > > ----------------------------------------------------------
> > >
> > > The source distributions is available at:
> > >
> > > ftp://ftp.ICSI.Berkeley.EDU/pub/real/davidj/feacat-1.01.tar.gz
> > >
> > > The following programs have been installed in
> > /u/drspeech/i586-linux/bin/
> > > and /u/drspeech/sun4-sunos5/bin/ at ICSI:
> > >
> > > feacat-1_01
> > > labcat-1_01
> > > linecat-1_01
> > > segpostsum-1_01
> > >
> > > These versions were linked with QuickNet v3_20.
> > >
> > > The default version of feacat has not been changed but these new
> > > versions will become the default in the near future if there are no
> > > problems.
> > >
> > > Cheers,
> > >
> > > David.
> > >
> > > --
> > > David Johnson, davidj@, (510) 666-2983
> > > International Computer Science Institute
> > >
> >
>

Wow, great! Why didn't I know this earlier? ;-)

Xuchen

________________________________________
From: icsi-speech-tools@yahoogroups.com
[mailto:icsi-speech-tools@yahoogroups.com] On Behalf Of Petr Fousek
Sent: Wednesday, June 13, 2007 6:40 PM
To: icsi-speech-tools@yahoogroups.com
Subject: [icsi-speech-tools] Re: feacalc doesn't support >2GB file?

> Looking in the NEWS file, I don't see any mention of testing with
> >=2GB pfiles. If it doesn't work, I'm guessing it just needs
> autoconf/header tweaking as all of the pfile writing is handled by
> the QuickNet routines.

We had that problem too at IDIAP. Guys from Brno did the neccessary
"tweaking". Check this page, search for "Note for SprachCore".

http://www.fit.vutbr.cz/research/groups/speech/sw/trapper.html.cs

Petr.

> Looking in the NEWS file, I don't see any mention of testing with
> >=2GB pfiles.  If it doesn't work, I'm guessing it just needs
> autoconf/header tweaking as all of the pfile writing is handled by
> the QuickNet routines.


We had that problem too at IDIAP. Guys from Brno did the neccessary
"tweaking". Check this page, search for "Note for SprachCore".

http://www.fit.vutbr.cz/research/groups/speech/sw/trapper.html.cs

Petr.

You are welcome.

Could you please give me a link about all the patches for gcc 4.1? I never
know there's such kind of patches.

Thanks.

Xuchen

________________________________________
From: icsi-speech-tools@yahoogroups.com
[mailto:icsi-speech-tools@yahoogroups.com] On Behalf Of Petr Fousek
Sent: Wednesday, June 13, 2007 6:23 PM
To: icsi-speech-tools@yahoogroups.com
Subject: [icsi-speech-tools] Re: feacat 1.01 released ---- dpwelib
dependence

Xuchen, your configure trick succeded; you were right also with gcc
version.

Meanwhile I found a patch for dpwelib-2002-10-23 (again from Olivier
Bornet) and the package compiled OK even with gcc 4.1.1. In summary,
after patching dpwelib-2002-10-23 and quicknet-v3.20, all packages
(QN3.20, feacat_1.01 and pfile_utils-v0.50) compiled OK with gcc 4.1.1.

Anyway, thanks for your suggestions.

Petr.

--- In icsi-speech-tools@yahoogroups.com, "yaoxuchen" <yaoxuchen@...>
wrote:
>
> The configure script error you mentioned is probably caused by
Tcl/Tk, if
> your linux box is quite new.
>
> Use the following script before configure:
>
> cp configure{,.orig} && sed "s/relid'/relid/" configure.orig > configure
>
> You may also need a lower gcc version (i.e. 3.4 or 3.2) to compile
dpwelib.
>
> Xuchen
>
> ________________________________________
> From: icsi-speech-tools@yahoogroups.com
> [mailto:icsi-speech-tools@yahoogroups.com] On Behalf Of Petr Fousek
> Sent: Friday, June 08, 2007 10:29 PM
> To: icsi-speech-tools@yahoogroups.com
> Subject: [icsi-speech-tools] Re: feacat 1.01 released ---- dpwelib
> dependence
>
> SHORT: Where one can obtain dpwelib sources?
>
> LONGER:
> - feacat 1.01 requires dpwelib and QuickNet3 libs to compile
>
> - QN version 3.20 compiled OK (thanks!), though some modified patches
> from Olivier Bornet for QN v3.11 were needed
>
> - the most recent dpwelib I could find is dpwelib-2006-04-19. However,
> this does neither compile on my current x64_64 linux box (configure
> script syntax error) nor on an old Celeron. Where can one get dpwelib
> which compiles so that feacat builds?
>
> Petr.
>
> --- In icsi-speech-tools@yahoogroups.com, David Johnson <davidj@>
> wrote:
> >
> > Version 1.01 of the feacat package has been released, tagged in the
> > CVS archive and installed in /u/drspeech/ at ICSI. feacat includes
> > several versatile utilities for munging feature files, probability
> > files, label files and line-based text files.
> >
> > This primary change in this releases is that disabling of "//" for
> > specifying parallel feature files by default, and the associated
> > "-usedoubleslash" option for enabling it. This brings it in line with
> > the version of feacat used at SRI.
> >
> > From the NEWS file:
> >
> > ----------------------------------------------------------
> > ====================
> > === feacat 1.01 ====
> > ====================
> >
> > - Added the "-usedoubleslash" option so the "//" syntax for parallel
> > feature files isn't enabled by default (it causes more trouble than
> > it's worth!)
> > - Fixed "ascii" format so it uses %g for output and added "asciif"
> which
> > uses the version 1.00 and before %f output.
> > - Added config.sub and config.guess to dist.
> > ----------------------------------------------------------
> >
> > The source distributions is available at:
> >
> > ftp://ftp.ICSI.Berkeley.EDU/pub/real/davidj/feacat-1.01.tar.gz
> >
> > The following programs have been installed in
> /u/drspeech/i586-linux/bin/
> > and /u/drspeech/sun4-sunos5/bin/ at ICSI:
> >
> > feacat-1_01
> > labcat-1_01
> > linecat-1_01
> > segpostsum-1_01
> >
> > These versions were linked with QuickNet v3_20.
> >
> > The default version of feacat has not been changed but these new
> > versions will become the default in the near future if there are no
> > problems.
> >
> > Cheers,
> >
> > David.
> >
> > --
> > David Johnson, davidj@, (510) 666-2983
> > International Computer Science Institute
> >
>

Xuchen, your configure trick succeded; you were right also with gcc
version.

Meanwhile I found a patch for dpwelib-2002-10-23 (again from Olivier
Bornet) and the package compiled OK even with gcc 4.1.1. In summary,
after patching dpwelib-2002-10-23 and quicknet-v3.20, all packages
(QN3.20, feacat_1.01 and pfile_utils-v0.50) compiled OK with gcc 4.1.1.

Anyway, thanks for your suggestions.

Petr.



--- In icsi-speech-tools@yahoogroups.com, "yaoxuchen" <yaoxuchen@...>
wrote:
>
> The configure script error you mentioned is probably caused by
Tcl/Tk, if
> your linux box is quite new.
>
> Use the following script before configure:
>
> cp configure{,.orig} && sed "s/relid'/relid/" configure.orig > configure
>
> You may also need a lower gcc version (i.e. 3.4 or 3.2) to compile
dpwelib.
>
> Xuchen
>
> ________________________________________
> From: icsi-speech-tools@yahoogroups.com
> [mailto:icsi-speech-tools@yahoogroups.com] On Behalf Of Petr Fousek
> Sent: Friday, June 08, 2007 10:29 PM
> To: icsi-speech-tools@yahoogroups.com
> Subject: [icsi-speech-tools] Re: feacat 1.01 released ---- dpwelib
> dependence
>
> SHORT: Where one can obtain dpwelib sources?
>
> LONGER:
> - feacat 1.01 requires dpwelib and QuickNet3 libs to compile
>
> - QN version 3.20 compiled OK (thanks!), though some modified patches
> from Olivier Bornet for QN v3.11 were needed
>
> - the most recent dpwelib I could find is dpwelib-2006-04-19. However,
> this does neither compile on my current x64_64 linux box (configure
> script syntax error) nor on an old Celeron. Where can one get dpwelib
> which compiles so that feacat builds?
>
> Petr.
>
> --- In icsi-speech-tools@yahoogroups.com, David Johnson <davidj@>
> wrote:
> >
> > Version 1.01 of the feacat package has been released, tagged in the
> > CVS archive and installed in /u/drspeech/ at ICSI. feacat includes
> > several versatile utilities for munging feature files, probability
> > files, label files and line-based text files.
> >
> > This primary change in this releases is that disabling of "//" for
> > specifying parallel feature files by default, and the associated
> > "-usedoubleslash" option for enabling it. This brings it in line with
> > the version of feacat used at SRI.
> >
> > From the NEWS file:
> >
> > ----------------------------------------------------------
> > ====================
> > === feacat 1.01 ====
> > ====================
> >
> > - Added the "-usedoubleslash" option so the "//" syntax for parallel
> > feature files isn't enabled by default (it causes more trouble than
> > it's worth!)
> > - Fixed "ascii" format so it uses %g for output and added "asciif"
> which
> > uses the version 1.00 and before %f output.
> > - Added config.sub and config.guess to dist.
> > ----------------------------------------------------------
> >
> > The source distributions is available at:
> >
> > ftp://ftp.ICSI.Berkeley.EDU/pub/real/davidj/feacat-1.01.tar.gz
> >
> > The following programs have been installed in
> /u/drspeech/i586-linux/bin/
> > and /u/drspeech/sun4-sunos5/bin/ at ICSI:
> >
> > feacat-1_01
> > labcat-1_01
> > linecat-1_01
> > segpostsum-1_01
> >
> > These versions were linked with QuickNet v3_20.
> >
> > The default version of feacat has not been changed but these new
> > versions will become the default in the near future if there are no
> > problems.
> >
> > Cheers,
> >
> > David.
> >
> > --
> > David Johnson, davidj@, (510) 666-2983
> > International Computer Science Institute
> >
>

>>>>> "yaoxuchen" == yaoxuchen  <yaoxuchen@...> writes:

     yaoxuchen> I use feacalc-0.90. And feacalc stops whenever the
     yaoxuchen> pfile is >= 2GB. I remembered that there maybe some
     yaoxuchen> historical issues that some tools from ICSI don't
     yaoxuchen> support >= 2GB file

Looking in the NEWS file, I don't see any mention of testing with
>=2GB pfiles.  If it doesn't work, I'm guessing it just needs
autoconf/header tweaking as all of the pfile writing is handled by the
QuickNet routines.

I've added a TODO file to the CVS archive with notes that this needs
to be checked out.

David.

     yaoxuchen> Could anybody please confirm?

     yaoxuchen> Xuchen






     yaoxuchen> Yahoo! Groups Links

>>>>> "yaoxuchen" == yaoxuchen  <yaoxuchen@...> writes:

     yaoxuchen> I use pfile_norm with option "-slf LENGTH_FILE" when do
     yaoxuchen> per speaker normalization.

     yaoxuchen> Whenever LENGTH_FILE is a little large, I got the
     yaoxuchen> following error:

     yaoxuchen> *** glibc detected *** free(): invalid next size
     yaoxuchen> (normal): 0x080781d8 *** Aborted

     yaoxuchen> After a little backtrace, the bug is found:

     yaoxuchen> pfile_norm.cc in pfile_utils_v0_50, line 416:

I've changed this in CVS.  Given that this isn't a particularly
serious issue, I'm not planning to do a release at this time.

David.

     yaoxuchen> fgets(buf+got, buflen, file);

     yaoxuchen> buf is buflen long, if we copy buflen chars to buf+got,
     yaoxuchen> then buf is out of boundary. And we get problem when
     yaoxuchen> free it.

     yaoxuchen> I believe it's better to write this way:

     yaoxuchen> fgets(buf+got, buflen-got, file);


     yaoxuchen> Regards,

     yaoxuchen> Xuchen





     yaoxuchen> Yahoo! Groups Links

I use pfile_norm with option "-slf LENGTH_FILE" when do per speaker
normalization.

Whenever LENGTH_FILE is a little large, I got the following error:

*** glibc detected *** free(): invalid next size (normal): 0x080781d8 ***
Aborted

After a little backtrace, the bug is found:

pfile_norm.cc in pfile_utils_v0_50, line 416:

fgets(buf+got, buflen, file);

buf is buflen long, if we copy buflen chars to buf+got, then buf is out of
boundary. And we get problem when free it.

I believe it's better to write this way:

fgets(buf+got, buflen-got, file);


Regards,

Xuchen

I use feacalc-0.90. And feacalc stops whenever the pfile is >= 2GB. I
remembered that there maybe some historical issues that some tools from ICSI
don't support >= 2GB file

Could anybody please confirm?

Xuchen

The configure script error you mentioned is probably caused by Tcl/Tk, if
your linux box is quite new.

Use the following script before configure:

cp configure{,.orig} && sed "s/relid'/relid/" configure.orig > configure

You may also need a lower gcc version (i.e. 3.4 or 3.2) to compile dpwelib.

Xuchen

________________________________________
From: icsi-speech-tools@yahoogroups.com
[mailto:icsi-speech-tools@yahoogroups.com] On Behalf Of Petr Fousek
Sent: Friday, June 08, 2007 10:29 PM
To: icsi-speech-tools@yahoogroups.com
Subject: [icsi-speech-tools] Re: feacat 1.01 released ---- dpwelib
dependence

SHORT: Where one can obtain dpwelib sources?

LONGER:
- feacat 1.01 requires dpwelib and QuickNet3 libs to compile

- QN version 3.20 compiled OK (thanks!), though some modified patches
from Olivier Bornet for QN v3.11 were needed

- the most recent dpwelib I could find is dpwelib-2006-04-19. However,
this does neither compile on my current x64_64 linux box (configure
script syntax error) nor on an old Celeron. Where can one get dpwelib
which compiles so that feacat builds?

Petr.

--- In icsi-speech-tools@yahoogroups.com, David Johnson <davidj@...>
wrote:
>
> Version 1.01 of the feacat package has been released, tagged in the
> CVS archive and installed in /u/drspeech/ at ICSI. feacat includes
> several versatile utilities for munging feature files, probability
> files, label files and line-based text files.
>
> This primary change in this releases is that disabling of "//" for
> specifying parallel feature files by default, and the associated
> "-usedoubleslash" option for enabling it. This brings it in line with
> the version of feacat used at SRI.
>
> From the NEWS file:
>
> ----------------------------------------------------------
> ====================
> === feacat 1.01 ====
> ====================
>
> - Added the "-usedoubleslash" option so the "//" syntax for parallel
> feature files isn't enabled by default (it causes more trouble than
> it's worth!)
> - Fixed "ascii" format so it uses %g for output and added "asciif"
which
> uses the version 1.00 and before %f output.
> - Added config.sub and config.guess to dist.
> ----------------------------------------------------------
>
> The source distributions is available at:
>
> ftp://ftp.ICSI.Berkeley.EDU/pub/real/davidj/feacat-1.01.tar.gz
>
> The following programs have been installed in
/u/drspeech/i586-linux/bin/
> and /u/drspeech/sun4-sunos5/bin/ at ICSI:
>
> feacat-1_01
> labcat-1_01
> linecat-1_01
> segpostsum-1_01
>
> These versions were linked with QuickNet v3_20.
>
> The default version of feacat has not been changed but these new
> versions will become the default in the near future if there are no
> problems.
>
> Cheers,
>
> David.
>
> --
> David Johnson, davidj@..., (510) 666-2983
> International Computer Science Institute
>

>>>>> "Petr" == Petr Fousek <p.fousek@...> writes:

     Petr> SHORT: Where one can obtain dpwelib sources?  LONGER: -
     Petr> feacat 1.01 requires dpwelib and QuickNet3 libs to compile

     Petr> - QN version 3.20 compiled OK (thanks!), though some
     Petr> modified patches from Olivier Bornet for QN v3.11 were
     Petr> needed

     Petr> - the most recent dpwelib I could find is
     Petr> dpwelib-2006-04-19. However, this does neither compile on my
     Petr> current x64_64 linux box (configure script syntax error) nor
     Petr> on an old Celeron. Where can one get dpwelib which compiles
     Petr> so that feacat builds?

You could try building dpwelib (and probably everything else) with
"-m32".  Note that the 64 bit pfile stuff doesn't require a 64 bit OS,
and you don't really lose anything running QuickNet in 32 bit mode.

David.

     Petr> Petr.




     Petr> --- In icsi-speech-tools@yahoogroups.com, David Johnson
     Petr> <davidj@...> wrote:
     >> Version 1.01 of the feacat package has been released, tagged in
     >> the CVS archive and installed in /u/drspeech/ at ICSI.  feacat
     >> includes several versatile utilities for munging feature files,
     >> probability files, label files and line-based text files.
     >>
     >> This primary change in this releases is that disabling of "//"
     >> for specifying parallel feature files by default, and the
     >> associated "-usedoubleslash" option for enabling it.  This
     >> brings it in line with the version of feacat used at SRI.
     >>
     >> From the NEWS file:
     >>
     >> ----------------------------------------------------------------
     >> ==================== === feacat 1.01 ==== ====================
     >>
     >> - Added the "-usedoubleslash" option so the "//" syntax for
     >> parallel feature files isn't enabled by default (it causes more
     >> trouble than it's worth!)  - Fixed "ascii" format so it uses %g
     >> for output and added "asciif"
     Petr> which
     >> uses the version 1.00 and before %f output.  - Added config.sub
     >> and config.guess to dist.
     >> ----------------------------------------------------------------
     >>
     >> The source distributions is available at:
     >>
     >> ftp://ftp.ICSI.Berkeley.EDU/pub/real/davidj/feacat-1.01.tar.gz
     >>
     >> The following programs have been installed in
     Petr> /u/drspeech/i586-linux/bin/
     >> and /u/drspeech/sun4-sunos5/bin/ at ICSI:
     >>
     >> feacat-1_01 labcat-1_01 linecat-1_01 segpostsum-1_01
     >>
     >> These versions were linked with QuickNet v3_20.
     >>
     >> The default version of feacat has not been changed but these
     >> new versions will become the default in the near future if
     >> there are no problems.
     >>
     >> Cheers,
     >>
     >> David.
     >>
     >> --
     >> David Johnson, davidj@..., (510) 666-2983 International
     >> Computer Science Institute
     >>





     Petr> Yahoo! Groups Links