[hts-users:03382] Re: Vowel duration modeling
On 11 Jul 2012, at 11:53, Ramil Məmmədov wrote:
> But when I add this 40 utterances to 2500-uterrance corpus (which doesn't have long vowels), there's no distinction between long and short vowels in synthesised speech.
> What can be problem in this case?
Did you add new questions for the clustering that differentiate between the long and short vowels? If not, they will simply be clustered together and share acoustic models.
If you did add such questions, then perhaps there are so few instances of the long vowels (relative to the short ones) that they get clustered together with the short vowels anyway. You can inspect the decision trees to see what questions were used, to find out if this is the case.
Simon
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
- References
-
- [hts-users:03381] Vowel duration modeling, Ramil Məmmədov