В Сбт, 10/10/2009 в 19:30 +0900, YounSung Park пишет: > Dear HTS maintainer, > > I am a student researching about the HTS based singing voice > synthesis. > > I think there is something I can't understand. > > I saw the paper about the parameter generation algorith. > > In the paper, the parameter generation algorithm has constraint in the > generated parameter O = WC. > > But actually, if the cepstrum vector generated from one state is long, > I can't find the relationship between the cepstrum sequence. > > ( I saw the plot of the 0-th and 2nd cepstrum vector and I find that > there's no discontinuity between the state transition. However after > the transition and 10 frame later, the generated cepstrum value is > flat..!! ) > > So I think that the parameter generation algorithm has some problem if > the duration length of one state is long. > > Thank you for reading this mail. > > Please answer my question and let me know the solution about that > problem. Dear YounSung, just to be precise you are probably referring to this article: http://hts.sp.nitech.ac.jp/?plugin=attach&refer=Publications&openfile=tokuda_icassp2000.pdf As for your question, there is no problem with spectrum being flat during the state, it's supposed to be so. This can be viewed as piecewise approximation for example. Feel free to ask other questions, but try to state them so we could understand.
Attachment:
signature.asc
Description: =?koi8-r?q?=FC=D4=C1?==?koi8-r?q?_=DE=C1=D3=D4=D8?= =?koi8-r?q?_=D3=CF=CF=C2=DD=C5=CE=C9=D1?= =?koi8-r?q?_=D0=CF=C4=D0=C9=D3=C1=CE=C1?= =?koi8-r?q?_=C3=C9=C6=D2=CF=D7=CF=CA?= =?koi8-r?q?_=D0=CF=C4=D0=C9=D3=D8=C0?=