[hts-users:01380] Re: questions about Tree-Based State Tying

[paminy <paminy@xxxxxxx>]

Thanks for your answer, Simon.
I still did not understand the "same" pdf.
Why they have the same PDF ? 
As there are three or five different states and also the pdfs in one frame ,I do not understand why they have the same pdf whith in so many different frames?

If using the same PDF,how dose the pdf's mean and variance caculate?
Dose it means the pdf's mean is the average of all  pdfs?
 
 

在2008-05-06，"Simon King" <Simon.King@xxxxxxxx> 写道：

paminy wrote:
> Hi,everyone
> paper "Tree-based State Tying for High Accuracy Acoustic Modeling" 
> said that " Let S be a set of HMM states and let L(S) be the log 
> likelihood of S generating the set of training frames F under the 
> assumping that all states in S are tied i.e.they share common mean and 
> variance and that transition probabilities can be ignored."
> I did not understand the sentence "they share common mean and 
> variance",how dose it means.

It means all the states in S use the same Gaussian PDF to generate their 
observations.

Simon

---

想买精品两居？先来看看实景二手房照片吧！