[hts-users:01939] Re: Up mixture Error

[Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>]

Hi, Shliu

Are all weights of the normal mixtures reduced to 0.0 in the embedded training?

Regards,
Keiichiro Oura

liuchangliang wrote:
> Hi:
>
>          I’m training a MSD-HMM. When I increased the number of the 
> mixtures using HHEd, an error happened.
>
> My initial MSD-HMM has two mixture, of which one is a normal mixture, 
> another is a zero-dimension mixture. After embedded training, the weight 
> of the normal mixture decreased to 0.00. Then HHEd treat it as a defunct 
> mixture and try to fix it. However, another mixture is a zero-dimension 
> mixture, so there’s no mixture being split. The error happened here.
>
>  
>
> Can anybody help me?
>
> Thanks
>
>  
>
> Chliu
>
> 2009/1/12