Error in applying G2P.

[Yash]

Hi,

I have been trying to do KWS for OOV words. I was able to train the confusion matrix and the g2p model. I am however getting an error when I am applying the g2p model. Below is the error message:

# cat stanford_cs/data/kws/g2p/wordlist.txt | g2p.py -V 0.95 --variants-number 10 --encoding utf-8 --model exp/g2p/g2p.model.final --apply - > stanford_cs/data/kws/g2p/output 

# Started at Fri Dec 11 23:33:33 EST 2015

#

Traceback (most recent call last):

  File "/data/ASR1/tools/G2P/g2p/g2p.py", line 38, in <module>

    import SequiturTool

  File "/data/ASR1/tools/G2P/g2p/SequiturTool.py", line 34, in <module>

    from sequitur import Sequitur, ModelTemplate, DefaultDiscountAdjuster, StaticDiscounts, FixedDiscounts, EagerDiscountAdjuster

  File "/data/ASR1/tools/G2P/g2p/sequitur.py", line 32, in <module>

    import sequitur_, SequenceModel, Minimization, misc

  File "/data/ASR1/tools/G2P/g2p/sequitur_.py", line 5, in <module>

    import _sequitur_

ImportError: /data/ASR1/tools/G2P/g2p/build/lib.linux-x86_64-2.4/_sequitur_.so: undefined symbol: Py_InitModule4

# Accounting: time=7 threads=1

# Ended (code 1) at Fri Dec 11 23:33:40 EST 2015, elapsed time 7 seconds

The funny part is, I was able to run this a few months ago. This issue seems like a python version mismatch to me. Can some one help me in figuring out what is wrong with my setup? Please let me know if any more information is required.

Thanks!

[Daniel Povey]

Sequitur for a long time has not been that easy to install as the original authors have not maintained.  I think recently Yenda (cc'd) modified our setup to use a fork of Sequitur that we control, that we can more easily apply patches and updates too.   I think it's best if you try to install it from scratch, and run any self-tests that it has, before we look into this further though.

Dan

--
You received this message because you are subscribed to the Google Groups "kaldi-help" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kaldi-help+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Jan Trmal]

Just guessing -- you've updated python during those few months and didn't recompile the sequitur.

Your best shot would be to try to recompile it (you can delete the directory and run the tools/install_sequitur.sh)

y.

[Yash]

Ok, let me try doing that. Thanks Yenda.

[Yash]

hi Yenda,

I am getting the following error message when I try to install sequitur:

[]$ bash install_sequitur.sh 

python setup.py build

running build

running build_py

copying sequitur_.py -> build/lib.linux-x86_64-2.7

running build_ext

building '_sequitur_' extension

swigging sequitur.i to sequitur_wrap.cpp

swig -python -c++ -shadow -o sequitur_wrap.cpp sequitur.i

gcc -pthread -fno-strict-aliasing -I/data/ASR1/tools/python27/include -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -fPIC -DMULTIGRAM_SIZE=2 -I/home/iyu/sfep/asr_noisemes/base/lib/python2.7/site-packages/numpy/core/include -I/data/ASR1/tools/python27/include/python2.7 -c sequitur_wrap.cpp -o build/temp.linux-x86_64-2.7/sequitur_wrap.o -DNPY_NO_DEPRECATED_API=NPY_1_7_API_VERSION

cc1plus: warning: command line option "-Wstrict-prototypes" is valid for Ada/C/ObjC but not for C++

sequitur_wrap.cpp: In function 'int SWIG_Python_ConvertFunctionPtr(PyObject*, void**, swig_type_info*)':

sequitur_wrap.cpp:2051: error: invalid conversion from 'const char*' to 'char*'

sequitur_wrap.cpp: In function 'void SWIG_Python_FixMethods(PyMethodDef*, swig_const_info*, swig_type_info**, swig_type_info**)':

sequitur_wrap.cpp:6664: error: invalid conversion from 'const char*' to 'char*'

error: command 'gcc' failed with exit status 1

make: *** [build] Error 1

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

Any ideas? am I using the incorrect version of gcc?

[Yash]

I think this should build against python 2.4 ? 

I assume it had a different kind of type checking... I am not sure though.

[Daniel Povey]

Yenda is in our server room at Hopkins where we have had an AC failure, and may not reply till to-morrow.

It could be a swig version issue-- what version is swig?

But I won't know what to do regardless.

Dan

[Yash]

Hi Dan,

SWIG is:

SWIG Version 1.3.29

[Jan Trmal]

Hi, I think the swig version is quite old -- can you please update it?

I don't think there would be a point in testing out different versions of gcc -- IIRC the python module build process always  calls the gcc the rest of the python was built with. Besides, I tested a bunch of various compilers when fixing the original sequitur and there wasn't any problem.

So I suspect the SWIG -- when I look in the generated C++ wrapper (sequitur_wrap.cpp) on my machine I see 

2420     /* here we get the method pointer for callbacks */

2421     const char *doc = (((PyCFunctionObject *)obj) -> m_ml -> ml_doc);

2422     const char *desc = doc ? strstr(doc, "swig_ptr: ") : 0;

so _if_ there was an incompatibility of the automatically generated SWIG code and recent gcc, it's been already fixed.

Just to be safe -- you have reasonably fresh kaldi, so that you are using the github sequitur repo, right?

y.