Spectral Transformation of Singing Vowels by Dynamic Frequency Warping

摘要：

There has been an ongoing demand for singing voice transformation. Most existing singing voice synthesis methods are focused on implementing the correct fundamental frequency (F0) trajectory. However, to produce high quality singing voice, the spectrum of the singing voice should also be carefully considered. In this research we look at the spectral differences of the singing voice at different fundamental frequency levels. To investigate the transformation of spectrum in singing transposition, we propose to use the Dynamic Frequency Warping (DFW) approach to learn the mapping of the spectrum curves of vowels at different music note level. From our experiment, we found that a natural sounding singing vowel can be generated by warping its spectrum according to a target vowel.

作者: Minghui Dong Paul Chan Ling Cen S. W. Lee

作者单位: Institute for Infocomm Research (I2R), A*Star, 1 Fusionopolis Way, Singapore 138632

会议类型: 国际会议

会议名称: 2011亚太信号与信息处理协会年度峰会(APSIPAASC 2011)

会议地点: 西安

会议语种:英文

页码: 1-4

在线出版日期: 2011-10-18（万方平台首次上网日期，不代表论文的发表时间）