[hts-users:01143] Re: HTS2.1beta-HVite

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

Hi,

Sacha Krstulovic wrote (2008/01/25 19:56):

> If one wanted to align some unseen acoustic observations
> with some HSMM states, would the conversion suggested
> by Simon with the use of HVite be the best way to achieve this,
> or would one have to hack HERest to export the affiliation
> between the states and the cumulated frames with best occupancy,
> or would there just be another better solution?

I think the best way is using the Viterbi algorithm for HSMMs.
We have proposed a decoding method for HSMMs based on weighted 
finite state transducers (WFSTs).
WFST is a very flexible framework.
By using WFSTs, we can easily develop a decoder for HSMMs.

Please refer to the following paper for details:

Keiichiro Oura, Heiga Zen, Yoshihiko Nankaku, Akinobu Lee, Keiichi Tokuda, 
Hidden semi-Markov model based speech recognition system using weighted finite-state transducer, 
Proc of ICASSP2006, pp.33-34, Toulouse, France, May 2006.

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
------------------------------------------------