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

[hts-users:00546] Re: ??: Re: Duration mode l?=


Hi,

Fang Li wrote:

Hi Hegiga

My name is Heiga, not Hegiga.

I read the corrections about duration model.In the training state,the probability of the state duration is calculated by its self-transition probability, isn't it?

Yes.

In synthesis stage,according to the paper "Duration Modeling For HMM-Based Speech Synthesis", the probability of the state duration should be calculated by a single gaussion distribution function which maximize the probability of the output probability of the observations.

Yes.

I found a function in HGen.c:SetStateDuration().I think it is used to set the state duration in synthesis stage.However it seems not maximizing the probability of the output probability of the observations or I have not found the right code for maximization.

Maximization of state output prob is done in HGen.c:ParamGen().

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://kt-lab.ics.nitech.ac.jp/~zen
------------------------------------------------


References
[hts-users:00545] 回复: Re: Duration mode l?=, Fang Li