eduzhai > Applied Sciences > Engineering >

Shouted Speech Compensation for Speaker Verification Robust to Vocal Effort Conditions

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: The performance of speaker verification systems degrades when vocal effortconditions between enrollment and test (e.g., shouted vs. normal speech) aredifferent. This is a potential situation in non-cooperative speakerverification tasks. In this paper, we present a study on different methods forlinear compensation of embeddings making use of Gaussian mixture models tocluster shouted and normal speech domains. These compensation techniques areborrowed from the area of robustness for automatic speech recognition and, inthis work, we apply them to compensate the mismatch between shouted and normalconditions in speaker verification. Before compensation, shouted condition isautomatically detected by means of logistic regression. The process iscomputationally light and it is performed in the back-end of an x-vectorsystem. Experimental results show that applying the proposed approach in thepresence of vocal effort mismatch yields up to 13.8 equal error rate relativeimprovement with respect to a system that applies neither shouted speechdetection nor compensation.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...