[hts-users:00365] Re: about my demo

["Nicholas Volk" <nvolk@xxxxxxxxxx>]

I have tested it, but currently I prefer a simple but proprietary Praat
script for doing this.

Essentially your problem is probably that pda fails to calculate f0
for some voiced sections. This could probably be alleviated by adjusting
various options, but I don't know what the optimal values would be.
Since there's propbably a plenty of voiced sections with calculated value
0 (meaning unvoiced) in your data, you end up having unvoiced states in
places they should not be (middle of vowels etc).
Changing the unvoiced/voiced threshold value might help
to remove/reduce the creakiness.

(I think that HZ said that current release candidate has support
for ESPS get_f0.)

regards,
  Nicholas

> Hi,
> I  have  used  HTS  to  generate chinese.
> But it has some  creaky voices.
>
> I use  PDA  from  Edinburgh to get f0  for  trianing   f0's   HMM.
>
> Does  PDA cause   the   creaky voices?
>
> Had  anyone used  PDA for HTS?
>
> thank you.
>
> Best regards
>