Menu
▾
▴
#18 online decoding sample shows error after updating newer revesion
Hello, recently i have updated kaldi revision and as i get updated to newer version, it says error when i do online decoding as below on your given testing database,
~/kaldi-online/src/online2bin/online2-wav-nnet2-latgen-faster --do-endpointing=false \
--online=false \
--config=nnet_a_gpu_online/conf/online_nnet2_decoding.conf \
--max-active=7000 --beam=15.0 --lattice-beam=6.0 \
--acoustic-scale=0.1 --word-symbol-table=graph/words.txt \
nnet_a_gpu_online/smbr_epoch2.mdl graph/HCLG.fst "ark:echo utterance-id1 utterance-id1|" "scp:echo utterance-id1 ENG_M.wav|" \
ark:/dev/null
so should i have to re-indexing now to solve it.
It's not enough to say that it shows an error. You need to paste the
output.
Dan
On Fri, Mar 6, 2015 at 1:34 PM, Ashish Dave badboys4life007@users.sf.net
wrote:
Related
Bugs:
#18here it is...
~/kaldi/src/online2bin/online2-wav-nnet2-latgen-faster --do-endpointing=false \ --online=false \ --config=nnet_a_gpu_online/conf/online_nnet2_decoding.conf \ --max-active=7000 --beam=15.0 --lattice-beam=6.0 \ --acoustic-scale=0.1 --word-symbol-table=graph/words.txt \ nnet_a_gpu_online/smbr_epoch2.mdl graph/HCLG.fst "ark:echo utterance-id1 utterance-id1|" "scp:echo utterance-id1 ENG_M.wav|" \ ark:/dev/null
vector::reverse
when i go in debugging mode, it get stopped in OpenFst Vector graph reading
Try to figure out from where "vector::reverse" is printed and why. Perhaps
it is an exception that gets thrown with that value and gets printed out by
Kaldi when it's caught in main. But I don't know why such an exception
would get thrown. Certainly Kaldi code would not do that, but perhaps
OpenFst code does it.. It could possibly be memory exhaustion, but
normally std::bad_alloc would be the exception.
Dan
On Fri, Mar 6, 2015 at 2:01 PM, Ashish Dave badboys4life007@users.sf.net
wrote:
Related
Bugs:
#18I just committed a bug-fix that could possibly affect this (to a bug
introduced 5 days ago), so please make sure you do any further testing with
newly updated source.
Dan
On Fri, Mar 6, 2015 at 2:28 PM, Daniel Povey danielpovey@users.sf.net
wrote:
Related
Bugs:
#18I have updated kaldi source to revision 4950 and rebuild it again but the problem still same. i used nnet2 sample database from http://kaldi-asr.org/downloads/build/2/sandbox/online/egs/fisher_english/s5 for decoding.
i have verified on debugger mode, error comes when it read file /graph/HCLG.fst and call method "VectorFstImpl *VectorFstImpl::Read(istream &strm, const FstReadOptions &opts)" on openfst/include/fst/vector-fst.h , it retuns nothing and shows logs as vector::reverse
errors comes in OpenFSt header file, not in kaldi source code. but before when i build on ubuntu 32bit OS, it was working completely and as i updated to 64bit OS, it doesnot work.
now i ll make a try to build a new model by myself on NNet2.
This has nothing to do with the nnet2 models and everything to do with the
way you compiled OpenFst and linked with it. Did you recompile OpenFst on
the 64 bit os? Did you run "make test" in src/?
You might want to run it in gdb (gdb --args (program) (args)) and type
"catch throw" and then "run" at the prompt, so if any exceptions are
thrown, you see where they are thrown from.
Dan
On Sat, Mar 7, 2015 at 5:44 AM, Ashish Dave badboys4life007@users.sf.net
wrote:
Related
Bugs:
#18hello Dam, yes i recompile OpenFst and even tried 1.2, 1.3.4 and 1.4 version of it and "make test" in src run SUCCESS
as per you said to run in gdb with "catch throw", here log comes
__cxa_throw() at 0x7ffff5198ea0
std::__throw_length_error() at 0x7ffff51ef5a6
std::vector<fst::arctpl\<fst::tropicalweighttpl\<float> >, std::allocator\<fst::arctpl\<fst::tropicalweighttpl\<float> > > >::reserve() at vector.tcc:70 0x70d529
fst::VectorFstBaseImpl<fst::vectorstate\<fst::arctpl\<fst::tropicalweighttpl\<float> > > >::ReserveArcs() at vector-fst.h:141 0x708773
fst::VectorFstImpl<fst::arctpl\<fst::tropicalweighttpl\<float> > >::Read() at vector-fst.h:351 0x70357b
fst::VectorFst<fst::arctpl\<fst::tropicalweighttpl\<float> > >::Read() at vector-fst.h:447 0x6ff9fe
fst::ReadFstKaldi() at fstext-utils-inl.h:1,276 0x6fc862
main() at online2-wav-nnet2-latgen-faster.cc:160 0x6f7a93</fst::arctpl\<fst::tropicalweighttpl\<float></fst::arctpl\<fst::tropicalweighttpl\<float></fst::vectorstate\<fst::arctpl\<fst::tropicalweighttpl\<float></fst::arctpl\<fst::tropicalweighttpl\<float></fst::arctpl\<fst::tropicalweighttpl\<float>
i used GCC 4.7 version for compiling and error comes on "/usr/include/c++/4.7/bits/vector.tcc" file.
If you can go up the stack and print the variables (using the "up" and
"print" commands) try to figure out what size it is trying to reserve (is
it negative?) and what it has just read. It could be that there is a
problem with the FST it is trying to read-- perhaps a corrupted file? Make
sure a differently compiled copy of Kaldi is able to read the same thing.
Dan
On Sun, Mar 8, 2015 at 3:27 AM, Ashish Dave badboys4life007@users.sf.net
wrote:
Related
Bugs:
#18Also, try using the commands fstcopy and fstrmepslocal on the same FST
input that it's crashing on. This will also tell me something about what
the problem might be.
Dan
On Sun, Mar 8, 2015 at 3:30 AM, Daniel Povey danielpovey@users.sf.net
wrote:
Related
Bugs:
#18hello Dan,
that error thrown by vector.tcc file in c++ 4.7 on line 70, i pasted some line below,
ifndef _VECTOR_TCC
define _VECTOR_TCC 1
namespace std _GLIBCXX_VISIBILITY(default)
{
_GLIBCXX_BEGIN_NAMESPACE_CONTAINER
template<typename _tp,="" typename="" _alloc="">
void
vector<_Tp, _Alloc>::
reserve(size_type __n)
{
if (__n > this->max_size())
__throw_length_error(__N("vector::reserve"));
if (this->capacity() < __n)</typename>
hello Dan,
problem solved by recompiling it again on g++ 4.8 version. as in debugging mode, i found error on g++ vector's code so i compile it on different version and it works on g++ 4.8 version.
thanks for yr guidance.
The only problem in g++'s vector code that I am aware of is a bug in
std::nth_element that is present in 4.8.2. So I'm skeptical of your
explanation- I'm wondering if it could be that the OpenFst compilation
picked up a different version of g++, and a different standard library,
than Kaldi compilation.
Dan
On Tue, Mar 10, 2015 at 9:06 AM, Ashish Dave badboys4life007@users.sf.net
wrote:
Related
Bugs:
#18