[hts-users:04621] Re: Trained Model can't pronounce correctly

[Ghost Gloomy <gloomyghostbit@xxxxxxxxx>]

Thanks, I will cheak each of them.

Sébastien Le Maguer <slemaguer@xxxxxxxxxxxxxxxxxxxx> 于2018年12月16日周日 下午11:56写道：

Hello,

here is a checklist which might help you:
  - is the mispronunciation following a pattern?
    - if yes:
       - check if the pronunciation dictionary correct
       - check if the question file consistent with the label files?
       - check if the mispronounced elements are in the same cluster:
         - check if the elements are in majority?
            - if yes, the model doesn't manage to capture it. Check the amount of data you have
            - if no, check the other elements and try to see if the "mispronunciation" is consistent with them.
         - if no, check the other elements and try to see if the "mispronunciation" is consistent with them.
     - if no:
       - check the aligment

Also, always checking the statistics of training corpus (nb segments per cluster, distributions,
...) is always a good start for understanding what is going on.

Kind regards,
Sébastien

-----------------------
On Sun, Dec 16 2018 (23:29 +08), Ghost Gloomy wrote:

> Hello, I created a model using HTS, but it can't pronounce correctly, what
> is causing this situation.
>
> Any advice?
>
>
> The model file:
> https://www.dropbox.com/s/stwn041ngsa6fd8/hy_voice.htsvoice?dl=0
> The Generated wav:
> https://www.dropbox.com/s/em7fzi3y2qgmlns/song070_f001_063.wav?dl=0
>
>                                           Many Thanks

--
Sébastien Le Maguer
Postdoctorate researcher - ADAPT centre - Trinity College Dublin
Co-chair of SYNSIG (https://synsig.org/index.php/Main_Page)