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

[hts-users:00068] Re: Building a Swedish HTS voice


Hi,

> Hello,
>
> I have begun work to create a Swedish HTS voice using the modified
> version of HTK. Since I don't have a Swedish diphone voice for Festival
> to begin with, I will begin by trying to create HMM:s for mel-cepstral
> parameters only. F0 and durations will be taken from an existing
> text-to-speech-system.

I think Johan Frid made some code for Festval which uses one
of the Swedish MBROLA voices:

http://www.ling.lu.se/persons/JohanF/festival/

The licence may or may not suit you.

I'd be surprised if KTH does not have a speech database.
It's another thing getting one's hand on it.
If you can access a speech database, everything is surprisingly
simple:

o you now have the raw files

o use the database  to build the appropriate *.utt files based on them
  with Festival. (Refer to Festival's clunits building stage where
  *.utt files are build.)

o use KTH's Wavesurfer/ESPS code to exctract the F0 contour.
  (the get F0 function, look at the wavesurfer code and see how
   Wavesurfer does it.)

o modify the question file to suit your needs better (nativize it)

o build

o enjoy

regards,
  Nicholas

References
[hts-users:00067] Building a Swedish HTS voice, Anders Lundgren