[Subject Prev][Subject Next][Thread Prev][Thread Next][Date Index][Thread Index]

[hts-users:02294] Re: parallel mode for HERest


It is not too late. I haven't focused on this probleme since the question but an answer is always welcomed.

Thank you,

Sébastien

Junichi Yamagishi a écrit :
Hi,

It might be too late to response answers.....


HTS version of HERest assume that the so-called dump file for
HMM (HER?.hmm.acc) is first and followed by the dump file for
duration (HER?.dur.acc).

But the perl script below simply combines all the dump fils as follows:

$command = "$command -p 0";
$command = "$command $ARGIN[$#ARGIN] $mmfi\/\*.acc";

This means the wrong order of the dump files, that is
HER1.dur.acc HER1.hmm.acc HER2.dur.acc HER2.hmm.acc ....

Regards,
Junichi


On 28 Jul 2009, at 11:08, Keiichiro Oura wrote:

Hi,

In the Bowon's script, HERest of HTK is defined.
It seems that you should use HERest of HTK instead of HTS.
Why don't you ask Bowon about it?

Regards,
Keiichiro Oura

Sebastien LE MAGUER wrote:
Hello,
I'm trying to "parallelize" the use of HERest using a modified version of the script of Bowon Lee(http://www.ifp.illinois.edu/~bowonlee/research/htk-pl/HERest.pl). The modification is the use of fork instead of qsub.
But the result of this is this error:
WARNING [-9999]  dump file for HMM include "hmm" in ../bin/HERest
WARNING [-9999] dump file for duration model include "dur" in ../bin/HERest
ERROR [+7191]  Infinite VaAcc!
FATAL ERROR - Terminating program ../bin/HERest
Does anyone have an idea ?
Thanks a lot,
Regards,
Sebastien






begin:vcard
fn;quoted-printable:S=C3=A9bastien Le Maguer
n;quoted-printable:Le Maguer;S=C3=A9bastien
org:;Equipe CORDIAL
adr:ENSSAT - 6 rue de Kerampont, BP 80518 ;;Bureau 203-I;Lannion Cedex;;22305 ;France
email;internet:Sebastien.Le_Maguer@xxxxxxxx
title:Doctorant
tel;work:02 96 46 91 29
url:http://www.irisa.fr/cordial/Members/slemague/
version:2.1
end:vcard


References
[hts-users:02101] parallel mode for HERest, Sebastien LE MAGUER
[hts-users:02134] Re: parallel mode for HERest, Keiichiro Oura
[hts-users:02293] Re: parallel mode for HERest, Junichi Yamagishi