[hts-users:03949] Re: Fwd: forced alignment regarding:-

[Sébastien Le Maguer <Sebastien.Le_maguer@xxxxxxxx>]

Hi,

the goal of the forced-alignment for no-silent GV step is to build new labels which are consistent to the trained models. The idea is to isolate silence for whom variance are, theoretically, are non. 

The two other steps leads to train (making gv) and generate global variance (making unseen). You can have more informations about global variance in these papers :
  - Toda, T., Black, A. W., & Tokuda, K. (2005, March). Spectral conversion based on maximum likelihood estimation considering global variance of converted parameter. In Proc. ICASSP (Vol. 1, pp. 9-12).
  - Tomoki, T. O. D. A., & Tokuda, K. (2007). A speech parameter generation algorithm considering global variance for HMM-based speech synthesis. IEICE TRANSACTIONS on Information and Systems, 90(5), 816-824.

Kind regards,
Sébastien


lintac.mp@xxxxxxxxx writes:

> Hi,
>
> I have few doubts in HTS.
>
> What is the significance of the steps,
>
>        forced alignment for no-silent GV
>        making global variance
>        making unseen models (GV)
>
>
> Thank you
> Actlin Jeeva

-- 
Dr. Sébastien Le Maguer - Research Engineer 
ENSSAT
6 rue de Kérampont
22300 Lannion
phone: (+33) 2 96 46 91 29
Mail: Sebastien.Le_maguer@xxxxxxxx
website : http://sites.google.com/site/seblemaguer