eduzhai > Applied Sciences > Engineering >

Phonological Features for 0-shot Multilingual Speech Synthesis

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Code-switching---the intra-utterance use of multiple languages---is prevalentacross the world. Within text-to-speech (TTS), multilingual models have beenfound to enable code-switching. By modifying the linguistic input tosequence-to-sequence TTS, we show that code-switching is possible for languagesunseen during training, even within monolingual models. We use a small set ofphonological features derived from the International Phonetic Alphabet (IPA),such as vowel height and frontness, consonant place and manner. This allows themodel topology to stay unchanged for different languages, and enables new,previously unseen feature combinations to be interpreted by the model. We showthat this allows us to generate intelligible, code-switched speech in a newlanguage at test time, including the approximation of sounds never seen intraining.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...