[hts-users:01925] Re: Parallel training by HERest

["Heiga ZEN (Byung Ha CHUN)" <heiga.zen@xxxxxxxxxxxxxxxxx>]

Hi,

tshlmail-hts@xxxxxxxxx wrote (2009/04/01 23:20):

> As far as I know, HERest supports parallel training as follows:
>
> # non-parallel:
> HERest -S trlist -I labs -H dir1/hmacs -M dir2 hmmlist    
>
> # parallel (equivalent to the above command)
> HERest -S trlist1 -I labs -H dir1/hmacs -M dir2 -p 1 hmmlist    # Part 1
> HERest -S trlist2 -I labs -H dir1/hmacs -M dir2 -p 2 hmmlist    # Part 2
> HERest                    -H dir1/hmacs -M dir2 -p 0 hmmlist dir2/*.acc    # Merging
>
> It is obvious that only the training set is split.

Yes.

> My current problem is that I have a huge model file (i.e. dir1/hmacs being huge) and insufficient memory. 
> So I am going to split the huge model file as well into small ones, such that each of them corresponds 
> to a partial training list (i.e. trlist*). In other words, I wonder if the following commands work:
>
> HERest -S trlist1 -I labs1 -H dir1/hmacs1 -M dir2 -p 1 hmmlist1    # Part 1
> HERest -S trlist2 -I labs2 -H dir1/hmacs2 -M dir2 -p 2 hmmlist2    # Part 2
> HERest                     -H dir1/hmacs  -M dir2 -p 0 hmmlist  dir2/*.acc    # Merging
>
> I tried this way today but got an error numbered +7150. I am not sure whether HERest supports splitting
> not only a training set but also a corresponding model file.

HTK doesn't support such split, so HTS also doesn't support it.

Best regards,

Heiga ZEN (Byung Ha CHUN)

--
Heiga ZEN (Byung Ha CHUN)
Speech Technology Group
Cambridge Research Lab
Toshiba Research Europe
phone: +44 1223 436975

______________________________________________________________________
This email has been scanned by the MessageLabs Email Security System.
For more information please visit http://www.messagelabs.com/email 
______________________________________________________________________