eduzhai > Applied Sciences > Engineering >

Disentangled speaker and nuisance attribute embedding for robust speaker verification

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 12 pages

Abstract: Over the recent years, various deep learning-based embedding methods havebeen proposed and have shown impressive performance in speaker verification.However, as in most of the classical embedding techniques, the deeplearning-based methods are known to suffer from severe performance degradationwhen dealing with speech samples with different conditions (e.g., recordingdevices, emotional states). In this paper, we propose a novel fully supervisedtraining method for extracting a speaker embedding vector disentangled from thevariability caused by the nuisance attributes. The proposed framework wascompared with the conventional deep learning-based embedding methods using theRSR2015 and VoxCeleb1 dataset. Experimental results show that the proposedapproach can extract speaker embeddings robust to channel and emotionalvariability.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...