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

[hts-users:02326] Re: About the duration model

Many thanks~

2009/12/23 Heiga ZEN (Byung Ha CHUN) <heiga.zen@xxxxxxxxxxxxxxxxx>

Xingchi Xian wrote:

I am wondering why the duration model is constructed with 5 dimensions instead of 5 states. These might cause some differences in the tree-based clustering.

Because they are highly correlated so we want to cluster them jointly.

In addition, can we reconstruct the duration model into 5 states with 1 dimension readily by editing some configure files?

No.  If you want to construct a duration decision tree for each state position separately, you can do it by modifying Config.pm like as follows:

 @cmp        = ('mgc','lf0');
- @dur        = ('dur');
+ @dur        = ('dur1','dur2','dur3','dur4','dur5');
 $ref{'cmp'} = \@cmp;

 %vflr = ('mgc' => '0.01', # variance floors
         'lf0' => '0.01',
-         'dur' => '0.01');
+         'dur1'=> '0.01);

I think HMGenS can work with this model, but I'm not sure whether hts_engine can handle duration model with state-position-dependent decision trees or not.


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 ______________________________________________________________________

[hts-users:02324] About the duration model, Xingchi Xian
[hts-users:02325] Re: About the duration model, Heiga ZEN (Byung Ha CHUN)