Re: [Kaldi-developers] online/sandbox errors

[Daniel P. <dp...@gm...>]

I think this is a bug in gcc version 4.8.1 that causes random crashes in
"nth_element".
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58800
This can't easily be worked around since nth_element is used in many
places.  But gcc has since been fixed.
Dan



On Tue, Apr 22, 2014 at 10:30 AM, Simon Klüpfel <sim...@gm...>wrote:

> Hi,
>
> I am experiencing yet another problem using the online decoder from the
> sandbox. I am using revision 3940.
>
> Everything up to the decoding stage works fine, and here it seems to throw
> segfaults rather randomly. while I managed to have a successful decoding
> run (see output attached), in most cases it just dies at some point.
>
> I attached the valgrind output. It shows once again 'Conditional jump or
> move depends on uninitialised value(s)', but now in the end also states
> that "the 'impossible' happened".
>
> I am not sure if there is another bug in the code, or if my system might
> be faulty.
>
> I am using MFCCs without transforms.
>
> Thanks,
>
> Simon
>
>
>
> On 04/14/2014 09:15 PM, Daniel Povey wrote:
>
>>
>> Oh-- I see.  I fixed it, thanks.
>> Dan
>>
>>
>>
>> On Mon, Apr 14, 2014 at 4:57 PM, Simon Klüpfel <sim...@gm...
>> <mailto:sim...@gm...>> wrote:
>>
>>     Hi Dan,
>>
>>     did you also have a look at the valgrind output? This is a different
>>     issue, I believe.
>>
>>     However, I am not sure if this is a bug in the C++ code, or a
>>     problem with my data or setup.
>>
>>     I had used the higher frequency for both training and testing
>>
>>     All the best,
>>
>>     Simon
>>
>>
>>     On 04/14/2014 07:21 PM, Daniel Povey wrote:
>>
>>         Thanks for the bug report-- fixed it.
>>         Regarding setting --sample-frequency=41000 in conf/mfcc.conf,
>>         this is OK
>>         as long as you do so in both train and test conditions, since
>>         features
>>         (as you probably know) are not comparable across different sample
>>         frequencies.
>>         Dan
>>
>>
>>
>>         On Mon, Apr 14, 2014 at 8:49 AM, Simon Klüpfel
>>         <sim...@gm... <mailto:sim...@gm...>
>>         <mailto:simon.kluepfel@gmail.__com
>>
>>         <mailto:sim...@gm...>>> wrote:
>>
>>              Hi,
>>
>>              there is a small bug in the prepare_online_decoding.sh
>>         script, I
>>              attached the diff.
>>
>>              I also experience a segfault. I am not sure if this is due
>>         to the
>>              data I am using, or a bug. I attached the output of the
>>         command, as
>>              well as the valgrind output reporting use of several
>>         uninitialized
>>              values.
>>
>>              My recordings are 441kHz, so I added
>>              --sample-frequency=44100
>>              to conf/mfcc.conf before running run_online_decoding.sh
>>
>>              Please tell me if you need any more info on my setup or
>>         anything else.
>>
>>              Simon
>>
>>
>>         ------------------------------__----------------------------
>> --__------------------
>>
>>              Learn Graph Databases - Download FREE O'Reilly Book
>>              "Graph Databases" is the definitive new guide to graph
>>         databases and
>>              their
>>              applications. Written by three acclaimed leaders in the
>> field,
>>              this first edition is now available. Download your free
>>         book today!
>>         http://p.sf.net/sfu/NeoTech
>>              _________________________________________________
>>              Kaldi-developers mailing list
>>         Kaldi-developers@lists.__sourceforge.net
>>         <mailto:Kal...@li...>
>>              <mailto:Kaldi-developers@__lists.sourceforge.net
>>         <mailto:Kal...@li...>>
>>         https://lists.sourceforge.net/__lists/listinfo/kaldi-__developers<
>> https://lists.sourceforge.net/lists/listinfo/kaldi-developers>
>>
>>
>>
>>
>>
>