eduzhai > Applied Sciences > Engineering >

Self-Expressing Autoencoders for Unsupervised Spoken Term Discovery

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Unsupervised spoken term discovery consists of two tasks: finding theacoustic segment boundaries and labeling acoustically similar segments with thesame labels. We perform segmentation based on the assumption that the framefeature vectors are more similar within a segment than across the segments.Therefore, for strong segmentation performance, it is crucial that the featuresrepresent the phonetic properties of a frame more than other factors ofvariability. We achieve this via a self-expressing autoencoder framework. Itconsists of a single encoder and two decoders with shared weights. The encoderprojects the input features into a latent representation. One of the decoderstries to reconstruct the input from these latent representations and the otherfrom the self-expressed version of them. We use the obtained features tosegment and cluster the speech data. We evaluate the performance of theproposed method in the Zero Resource 2020 challenge unit discovery task. Theproposed system consistently outperforms the baseline, demonstrating theusefulness of the method in learning representations.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×