eduzhai > Applied Sciences > Engineering >

Dynamic Noise Embedding Noise Aware Training and Adaptation for Speech Enhancement

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 8 pages

Abstract: Estimating noise information exactly is crucial for noise aware training inspeech applications including speech enhancement (SE) which is our focus inthis paper. To estimate noise-only frames, we employ voice activity detection(VAD) to detect non-speech frames by applying optimal threshold on speechposterior. Here, the non-speech frames can be regarded as noise-only frames innoisy signal. These estimated frames are used to extract noise embedding, nameddynamic noise embedding (DNE), which is useful for an SE module to capture thecharacteristic of background noise. The DNE is extracted by a simple neuralnetwork, and the SE module with the DNE can be jointly trained to be adaptiveto the environment. Experiments are conducted on TIMIT dataset forsingle-channel denoising task and U-Net is used as a backbone SE module.Experimental results show that the DNE plays an important role in the SE moduleby increasing the quality and the intelligibility of corrupted signal even ifthe noise is non-stationary and unseen in training. In addition, we demonstratethat the DNE can be flexibly applied to other neural network-based SE modules.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×