[hts-users:02574] Re: About the parameter generation algorithm

[那兴宇 <nxy-yzqs@xxxxxxx>]

Hi, Zen
Thanks very much for your reply.

I am wondering if it is possible to solve the matrix formula frame by frame to get a whole dimension of feature vector in the loop of one frame. I mean obtaining a full C(t) instead of getting one dimension of all the C(t)s and then the next.

This is for the purpose of Real-Time synthesis because the synthesizer has a very long lag when processing a long sentence. And, the covariance matrices are diagonal.

If there is such algorithm, would you please offer me a document describing that? Thanks a lot!

Sorry to bother you.

>Hi,
>
>(02/08/2010 14:53), 那兴宇 wrote:
>
>> I am reading the source codes of hts_engine and I met up with this question.
>> Why does the numerical solution of WUW*C=WUM work in the order of time length instead of vector length?
>> Does the decomposition and substitution method work out that way?
>
>Because hts_engine assumes diagonal covariance matrices, we can perform the decomposition and
>substitution for each dimension separately.  So WUW and WUM work in the order of time length (T)
>instead of vector length (M*T).  This is more memory efficient than solving all dimensions
>simultaneously.
>
>HMGenS supports both diagonal & full covariance matrices.  If you use HMGenS with full covariance
>matrices, WUW and WUM work in the order of vector length.
>
>Regards,
>
>Heiga ZEN (Byung Ha CHUN)
>
>======================================================================
>NOTE: The information in this email and any attachments may be
>confidential and/or legally privileged. This message may be read, copied
>and used only by the intended recipient. If you are not the intended
>recipient, please destroy this message, delete any copies held on your
>system and notify the sender immediately.
>
>Toshiba Research Europe Limited, registered in England and Wales (2519556).
>Registered Office: 208 Cambridge Science Park, Milton Road, Cambridge
>CB4 0GZ, England. Web: http://www.toshiba-europe.com/research/crl
>======================================================================
>
>
>______________________________________________________________________
>This email has been scanned by the MessageLabs Email Security System.
>For more information please visit http://www.messagelabs.com/email 
>______________________________________________________________________
>

---

网易邮箱，没有垃圾邮件的邮箱。