[hts-users:01404] Re: Help on HEREST

["Heiga ZEN (Byung Ha CHUN)" <zen@xxxxxxxxxxxxxxx>]

Hi,

Qian Xiaojun wrote (2008/05/21 15:11):

> Hi Dr Heiga,
>    Thanks very much. I wonder what is a difference between the two cases 
> below:
> 
> 1. HEREST -e by using the model alignment
> 2. HREST for full context labels individually.

For example,

00000000 02000000 a
02000000 04500000 b

if durations of phoneme "a" and "b" are 4 and 5 frames, respectively, and each HMM has 5 emission states with no skip structure, 
HInit and HRest do not use this segment to estimate model parameters of "a".
However, HERest with -e option automatically relax the model boundary constraints 
and use observations in this segment to estimate model parameters of "a".
Please refer to HFB.c:SetAlignBeam() for details.

Regards,

Heiga ZEN (Byung Ha CHUN)

-- 
------------------------------------------------
 Heiga ZEN     (in Japanese pronunciation)
 Byung Ha CHUN (in Korean pronunciation)

 Department of Computer Science and Engineering
 Nagoya Institute of Technology
 Gokiso-cho, Showa-ku, Nagoya 466-8555 Japan

 http://www.sp.nitech.ac.jp/~zen
------------------------------------------------