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

[hts-users:00835] Re: about fullcontext model clustering


Hi :
        Yes, I see.                                                    
                                                        Thank you. 
----- Original Message ----- 
From: "Simon King" <Simon.King@xxxxxxxx>
To: <hts-users@xxxxxxxxxxxxxxx>
Sent: Thursday, October 11, 2007 6:05 PM
Subject: [hts-users:00834] Re: about fullcontext model clustering


> Heng Lu wrote:
>> Hi   :
>>        I have changed the command line to 
>>        TB 000  f0_s_ {*.state[2,3,4,5,6].tream[2-4]}
>>       But there is an error:            
>>           "ERROR [+2663]    CreateTree: attempt to cluster different states"
> 
> OK - this limitation is hardwired into HHEd, so you have to hack a bit 
> of code to make it possible. I did do this a long time ago, and I don't 
> think it was hard, but I don't have that code anymore. It may even be as 
> simple as converting the error to a warning.
> 
> If hacking the code is to hard, you could do a bit of surgery on the MMF 
> to make a dummy set of 1-state models containing the states you want to 
> cluster, then do the clustering, then reconstruct your multi-state 
> models from the clustered states. This is not too painful, because it's 
> all in terms of macros.
> 
> Simon
> 
> 
>

References
[hts-users:00831] about fullcontext model clustering, Heng Lu
[hts-users:00833] Re: about fullcontext model clustering, Heng Lu