According to the aforementioned, we summarize the design rules as follows:
1. Cover all syllables (in Mandarin, there are about 411 base-tone syllables.)
2. Cover all coarticulation between any two syllables.
3. Cover as many various pitch levels and durations for each syllable as possible.
With the above rules, we have three corpora: single-syllable-based corpus (SSC), coarticulation-based corpus (CC), and songs-based corpus (SC).
(Lin, Cheng-Yuan, Tzu-Ying Lin, and J-S. Roger Jang. "A corpus-based singing voice synthesis system for Mandarin Chinese." Proceedings of the 13th annual ACM international conference on Multimedia. ACM, 2005.)
根据上述提到的内容,我们将设计规则总结如下:
1. 覆盖全部的音节(在普通话中,共有大约411个平声的音节)
2. 覆盖全部的双音节间的协同发音
3. 为每个音节覆盖尽可能多的音高和时长组合
按照上述规则,我们设计出三个语料库:基于单音节的语料库(SSC)、基于协同发音的语料库(CC)、和基于歌曲的语料库(SC)。
HideshimaIori 发表于 2014/11/16 10:32
楼主一定非常恨UTAU及其使用者【。
既然这个软件及其用家都像你说的那么烂那干嘛不开发好自己的软件再来
RU ...
HideshimaIori 发表于 2014/11/16 10:32
楼主一定非常恨UTAU及其使用者【。
既然这个软件及其用家都像你说的那么烂那干嘛不开发好自己的软件再来
RU ...
欢迎光临 iVocaloid论坛 (http://bbs.ivocaloid.com/) | Powered by Discuz! X2 |