eduzhai > Applied Sciences > Engineering >

DCCRN Deep Complex Convolution Recurrent Network for Phase-Aware Speech Enhancement

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Speech enhancement has benefited from the success of deep learning in termsof intelligibility and perceptual quality. Conventional time-frequency (TF)domain methods focus on predicting TF-masks or speech spectrum, via a naiveconvolution neural network (CNN) or recurrent neural network (RNN). Some recentstudies use complex-valued spectrogram as a training target but train in areal-valued network, predicting the magnitude and phase component or real andimaginary part, respectively. Particularly, convolution recurrent network (CRN)integrates a convolutional encoder-decoder (CED) structure and long short-termmemory (LSTM), which has been proven to be helpful for complex targets. Inorder to train the complex target more effectively, in this paper, we design anew network structure simulating the complex-valued operation, called DeepComplex Convolution Recurrent Network (DCCRN), where both CNN and RNNstructures can handle complex-valued operation. The proposed DCCRN models arevery competitive over other previous networks, either on objective orsubjective metric. With only 3.7M parameters, our DCCRN models submitted to theInterspeech 2020 Deep Noise Suppression (DNS) challenge ranked first for thereal-time-track and second for the non-real-time track in terms of Mean OpinionScore (MOS).

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×