Simon having trouble with Julius

[Peter Grasch]

On Tuesday, June 17, 2014 09:57:42 AM Shantanu Tushar Jha wrote:
> > Ah. So you have the HTK installed?
> 
> Yep, from Ubuntu 14.04 packages. It worked quite good for a day ;)
Packages? Oh god, are there binary packages? There really, really shouldn't 
be...

I can't find any on packages.ubuntu.com. Where did you get yours from?

> > Adaption is always a bit tricky and the support in Simon isn't that great
> > to
> > be honest. One thing that's missing, for example, is to ensure that the
> > adaption set is sufficiently large for it to make sense to be used (where
> > "sufficiently large" is really hard to judge programmatically), as
> > adapting
> > using a small data set usually produces fairly stupid models.
> 
> Ah ok, wasn't too stupid for me actually, I'd say 8/10 times it did what I
> said ;)
If you don't have any adaption data yet, it uses the static base model even in 
"adapted" mode. But hey, if you have a limited vocab, reaching something like 
80 % accuracy really isn't that hard. I'm just saying that adaption usually 
decreases overall system performance if the adaption corpus is overly small.

> > Still, that should never cause your model to not have a tiedlist (which is
> > just copied from the base model).
> > Could you check if "base.sbm" (in the same directory as active.sbm) has
> > it?
> > Just for the record: You didn't get an error dialog that said "Could not
> > switch to new model.", right?
> 
> This is what I have-
> 
> shantanu at shaan-desktop:~$ find  . -name active.sbm
> ./.kde/share/apps/simon/model/active.sbm
> ./.kde/share/apps/simond/models/default/active/active.sbm
> 
> shantanu at shaan-desktop:~$ find  . -name base.sbm
> ./.kde/share/apps/simond/models/default/active/base.sbm
Yes, I'd be interested in the base model that the server has (simond).

> > How many adaption samples did you record?
> 
> I recorded all samples in Window Management twice, the Firefox one once.
Okay.
Could you please try and run one more training?
The additional data will trigger a model rebuild. It would be really great if 
the problem would persist but my initial guess is that's you've hit a race 
condition in the model cache / compilation / synchronization.

Best regards,
Peter