[hts-users:00775] 答复: [hts-users:00772] Re: Questions on training and flat pitch pattern

["lixiulin" <lixiulin@xxxxxxxxxxxxxxxxxxxxx>]

Hi, 

The error occurs when "copy monophone models to fullcontext ones". Related information is:

12,000 sentences are used, and there are totally 850,000 fullcontexts. We noticed that the ram and swap(3G in total) are used up when copying monophone models to fullcontext ones, and then failed to malloc memory . As a result, it can't finish this step or go to the following tree-based clustering procedures. We ever trained the model using 1500 sentences, and it runs well. The problem may be how to reduce the memory consumption in clone procedure. 

"?" is not used in fullcontext. what the use of it?


Best regards,

Sillon Lee



-----原始邮件-----
发件人: Junichi Yamagishi [mailto:jyamagis@xxxxxxxxxxxx]
发送时间: 2007年8月8日 20:31
收件人: hts-users@xxxxxxxxxxxxxxx
抄送: Junichi Yamagishi
主题: [hts-users:00772] Re: Questions on training and flat pitch pattern


Hi,

On 2007/08/08, at 6:41, lixiulin wrote:
> 1. after -r option is added to HHED, it still doesn't work. The  
> error messages are as follows:
>
> ====================================================================== 
> ==========
> Start copying monophone mmf to fullcontext one at 三  8 
> 月  8 10:15:51 CST 2007
> ====================================================================== 
> ==========
>
> /home/kangyg/HTS/tools/HTK-3.4/bin/HHEd -A -B -C /home/kangyg/HTS/ 
> HTS-AOZORAR/configs/trn.cnf -D -T 1 -p -i -r -H /home/kangyg/HTS/ 
> HTS-AOZORAR/models/qst001/ver1/cmp/monophone.mmf -w /home/kangyg/ 
> HTS/HTS-AOZORAR/models/qst001/ver1/cmp/fullcontext.mmf /home/kangyg/ 
> HTS/HTS-AOZORAR/edfiles/qst001/ver1/cmp/m2f.hed /home/kangyg/HTS/ 
> HTS-AOZORAR/data/lists/mono.list
>
> HTK Configuration Parameters[3]
>   Module/Tool     Parameter                  Value
> #                 NATURALWRITEORDER              TRUE
> #                 NATURALREADORDER              TRUE
> #                 APPLYVFLOOR                 TRUE
>
> HHEd
>  39/39 Models Loaded [7 states max, 2 mixes max]
>
> // copy monophone models to fullcontext ones
> CL /home/kangyg/HTS/HTS-AOZORAR/data/lists/full.list
>  Cloning current hmms to produce new set
> Error in /home/kangyg/HTS/tools/HTK-3.4/bin/HHEd   -A -B -C /home/ 
> kangyg/HTS/HTS-AOZORAR/configs/trn.cnf -D -T 1 -p -i -r -H /home/ 
> kangyg/HTS/HTS-AOZORAR/models/qst001/ver1/cmp/monophone.mmf -w / 
> home/kangyg/HTS/HTS-AOZORAR/models/qst001/ver1/cmp/fullcontext.mmf / 
> home/kangyg/HTS/HTS-AOZORAR/edfiles/qst001/ver1/cmp/m2f.hed /home/ 
> kangyg/HTS/HTS-AOZORAR/data/lists/mono.list
Unfortunately, we can not understand your problem from these error  
message.
Please send us detailed error message.

BTW, do your full.list include "?" symbol?



Then, please use -r options in the the following tree-based  
clustering procedures.

Training.pl

# HHEd (tree-based context clustering (cmp))
if ($CXCL1) {
    print_time("tree-based context clustering (cmp)");

    shell("cp $fullmmf{'cmp'} $clusmmf{'cmp'}");

    # tree-based clustering
    $footer = "";
    foreach $type (@cmp) {
       if ($strw{$type}>0.0) {
          make_edfile_state($type);
          shell("$HHEd{'trn'} -r -H $clusmmf{'cmp'} $mdl{$type} -w  
$clusmmf{'cmp'} $cxc{$type} $lst{'ful'}");
          $footer .= "_$type";
          shell("gzip -c $clusmmf{'cmp'} > $clusmmf{'cmp'}$footer.gz");
       }
    }
}


# HHEd (tree-based context clustering (cmp))
if ($CXCL2) {
    print_time("tree-based context clustering (cmp)");

    # tree-based clustering
    $set = 'cmp';

    shell("cp $untymmf{$set} $reclmmf{$set}");

    $footer = "";
    foreach $type (@{$ref{$set}}) {
       make_edfile_state($type);
       shell("$HHEd{'trn'} -r -H $reclmmf{$set} $mdl{$type} -w  
$reclmmf{$set} $cxc{$type} $lst{'ful'}");

       $footer .= "_$type";
       shell("gzip -c $reclmmf{$set} > $reclmmf{$set}$footer.gz");
    }
    shell("gzip -c $reclmmf{$set} > $reclmmf{$set}.noembedded.gz");
}

# HHEd (tree-based context clustering (dur))
if ($CXCL3) {
    print_time("tree-based context clustering (dur)");

    # tree-based clustering
    $set = 'dur';

    shell("cp $untymmf{$set} $reclmmf{$set}");
    convstats();

    $footer = "";
    foreach $type (@{$ref{$set}}) {
       make_edfile_state($type);
       shell("$HHEd{'trn'} -r -H $reclmmf{$set} $mdl{$type} -w  
$reclmmf{$set} $cxc{$type} $lst{'ful'}");

       $footer .= "_$type";
       shell("gzip -c $reclmmf{$set} > $reclmmf{$set}$footer.gz");
    }
    shell("gzip -c $reclmmf{$set} > $reclmmf{$set}.noembedded.gz");
}