eduzhai > Applied Sciences > Engineering >

DurIAN-SC Duration Informed Attention Network based Singing Voice Conversion System

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Singing voice conversion is converting the timbre in the source singing tothe target speaker s voice while keeping singing content the same. However,singing data for target speaker is much more difficult to collect compared withnormal speech this http URL this paper, we introduce a singing voice conversionalgorithm that is capable of generating high quality target speaker s singingusing only his her normal speech data. First, we manage to integrate thetraining and conversion process of speech and singing into one framework byunifying the features used in standard speech synthesis system and singingsynthesis system. In this way, normal speech data can also contribute tosinging voice conversion training, making the singing voice conversion systemmore robust especially when the singing database is small.Moreover, in order toachieve one-shot singing voice conversion, a speaker embedding module isdeveloped using both speech and singing data, which provides target speakeridentify information during conversion. Experiments indicate proposed singconversion system can convert source singing to target speaker s high-qualitysinging with only 20 seconds of target speaker s enrollment speech data.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...