eduzhai > Applied Sciences > Engineering >

Efficient neural speech synthesis for low-resource languages through multilingual modeling

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Recent advances in neural TTS have led to models that can producehigh-quality synthetic speech. However, these models typically require largeamounts of training data, which can make it costly to produce a new voice withthe desired quality. Although multi-speaker modeling can reduce the datarequirements necessary for a new voice, this approach is usually not viable formany low-resource languages for which abundant multi-speaker data is notavailable. In this paper, we therefore investigated to what extent multilingualmulti-speaker modeling can be an alternative to monolingual multi-speakermodeling, and explored how data from foreign languages may best be combinedwith low-resource language data. We found that multilingual modeling canincrease the naturalness of low-resource language speech, showed thatmultilingual models can produce speech with a naturalness comparable tomonolingual multi-speaker models, and saw that the target language naturalnesswas affected by the strategy used to add foreign language data.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×