eduzhai > Applied Sciences > Engineering >

On Loss Functions and Recurrency Training for GAN-based Speech Enhancement Systems

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Recent work has shown that it is feasible to use generative adversarialnetworks (GANs) for speech enhancement, however, these approaches have not beencompared to state-of-the-art (SOTA) non GAN-based approaches. Additionally,many loss functions have been proposed for GAN-based approaches, but they havenot been adequately compared. In this study, we propose novel convolutionalrecurrent GAN (CRGAN) architectures for speech enhancement. Multiple lossfunctions are adopted to enable direct comparisons to other GAN-based systems.The benefits of including recurrent layers are also explored. Our results showthat the proposed CRGAN model outperforms the SOTA GAN-based models using thesame loss functions and it outperforms other non-GAN based systems, indicatingthe benefits of using a GAN for speech enhancement. Overall, the CRGAN modelthat combines an objective metric loss function with the mean squared error(MSE) provides the best performance over comparison approaches across manyevaluation metrics.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×