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

[hts-users:04443] Re: Corrupt Synthesis for Chichewa Language


Hello,

your problem has multiple possible causes:
  - the extraction of the coefficient failed : try to check the extracted coefficients by regenerating the speech from them or by hand checking them
  - the training failed: check the models (you can dump them in an ASCII using the hlist.conf, search on the mailing list, there is an issue about that)

Kind regards,
Sébastien

On Wed, Jan 11 2017 (05:09), Jeremiah Chienda <jeremiahchienda@xxxxxxxxx> wrote:

> Dear All
>
> I am using theHTS-demo_CMU-ARCTIC-SLT to train a model for my native language of Chichewa. I don’t have full context labels so I have gone ahead with the monpophone labels I generated using forced alignment. However, I’m getting the following error during my training:
>
> x2x : warning: input data is over the range of type 'short’!
>
> The resulting ‘synthesized’ file is corrupt and sounds like a continuous stream of noise. Does anyone understand what the problem could be, and more importantly, how to solve it?
>
> Regards
>
> Jeremiah Chienda


--
Dr. Sébastien Le Maguer
Postdoctorate researcher

Saarland University
Campus C7.4 - room 2.03
D-66123 Saarbrücken
Germany

phone : +49-681-302-70030
Mail: slemaguer@xxxxxxxxxxxxxxxxxxxx
website :  http://www.coli.uni-saarland.de/~slemaguer/

References
[hts-users:04442] Corrupt Synthesis for Chichewa Language, Jeremiah Chienda