eduzhai > Applied Sciences > Engineering >

Data augmentation and loss normalization for deep noise suppression

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 8 pages

Abstract: Speech enhancement using neural networks is recently receiving largeattention in research and being integrated in commercial devices andapplications. In this work, we investigate data augmentation techniques forsupervised deep learning-based speech enhancement. We show that not onlyaugmenting SNR values to a broader range and a continuous distribution helps toregularize training, but also augmenting the spectral and dynamic leveldiversity. However, to not degrade training by level augmentation, we propose amodification to signal-based loss functions by applying sequence levelnormalization. We show in experiments that this normalization overcomes thedegradation caused by training on sequences with imbalanced signal levels, whenusing a level-dependent loss function.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×