[hts-users:03585] Re: HSMMAlign memory requirements

[Dietmar Schabus <schabus@xxxxxx>]

Ah yes, it's much faster and requires much less memory now.

Thank you!


On 2013-01-21 18:19, Keiichiro Oura wrote:
> Hi,
>
> Could you try to use -t option (beam width) to reduce memory requirements.
> For examples,
>
> % HSMMAlign -t 1000 1000 100000 -A -T ......
>
> Regards,
> Keiichiro Oura
>
>
>
> 2013/1/22 Dietmar Schabus<schabus@xxxxxx>:
> > Hello,
> >
> > I am using HSMMAlign of HTS 2.2 for forced alignment. It works great, except
> > for larger files: when I call HSMMAlign like
> >
> > HSMMAlign \
> >     -A \
> >     -T 2 \
> >     -C data/ac_sd/dsc/config/general.conf \
> >     -m hsmmalignments/ac_sd/dsc \
> >     -I data/ac_sd/dsc/list/full.mlf \
> >     -H data/ac_sd/dsc/models/1/clustered.cmp.mmf \
> >     -N data/ac_sd/dsc/models/1/clustered.dur.mmf \
> >     data/ac_sd/dsc/models/1/context.cmp.list \
> >     data/ac_sd/dsc/models/1/context.dur.list \
> >     data/ac_sd/dsc/Analysis/obs/dsc_butter_004.cmp
> >
> > then it fills up all of my 4 GiB of RAM, then also all of my 4 GiB of swap
> > space, and then gets killed by the operating system.
> >
> > dsc_butter_004.cmp is 1.4 MiB, the original sound file was a bit more than 9
> > seconds long.
> >
> > Is there a way to reduce the memory requirements? Maybe the -t parameter?
> > But what exactly is pruned here, and what are adequate values for -t?
> > Maybe HTS 2.3 beta has some changes in this regard?
> >
> > Thanks a lot for your help,
> > Dietmar
> >
> > --
> > ----------------------------------------------------------------------
> > Dipl.-Ing. Dietmar Schabus | Researcher
> > phone +43 1 5052830-48 | fax -99 | schabus@xxxxxx |
> > http://userver.ftw.at/~schabus/
> >
> > FTW Telecommunications Research Center Vienna
> > Donau-City-Straße 1/3 | 1220 Vienna | Austria | www.ftw.at