eduzhai > Applied Sciences > Engineering >

Emotion Profile Refinery for Speech Emotion Classification

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Human emotions are inherently ambiguous and impure. When designing systems toanticipate human emotions based on speech, the lack of emotional purity must beconsidered. However, most of the current methods for speech emotionclassification rest on the consensus, e.g., one single hard label for anutterance. This labeling principle imposes challenges for system performanceconsidering emotional impurity. In this paper, we recommend the use ofemotional profiles (EPs), which provides a time series of segment-level softlabels to capture the subtle blends of emotional cues present across a specificspeech utterance. We further propose the emotion profile refinery (EPR), aniterative procedure to update EPs. The EPR method produces soft,dynamically-generated, multiple probabilistic class labels during successivestages of refinement, which results in significant improvements in the modelaccuracy. Experiments on three well-known emotion corpora show noticeable gainusing the proposed method.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×