ePrints

Convolutional fusion network for monaural speech enhancement

Lookup NU author(s): Yang Xian, Dr Mohsen Naqvi

Downloads

• Accepted version [.pdf]

Licence

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND).

Abstract

Convolutional neural network (CNN) based methods, such as the convolutional encoder-decoder net-work, oer state-of-the-art results in monaural speech enhancement. In the conventional encoder-decoder network, large kernel size is often used to enhance the model capacity, which, however, re-sults in low parameter eciency. This could be addressed by using group convolution, as in AlexNet,where group convolutions are performed in parallel in each layer, before their outputs are concatenated.However, with the simple concatenation, the inter-channel dependency information may be lost. Toaddress this, the Shue network re-arranges the outputs of each group before concatenating them, bytaking part of the whole input sequence as the input to each group of convolution. In this work, wepropose a new convolutional fusion network (CFN) for monaural speech enhancement by improvingmodel performance, inter-channel dependency, information reuse and parameter eciency. First,a new group convolutional fusion unit (GCFU) consisting of the standard and depth-wise separableCNN is used to reconstruct the signal. Second, the whole input sequence (full information) is fedsimultaneously to two convolution networks in parallel, and their outputs are re-arranged (shued)and then concatenated, in order to exploit the inter-channel dependency within the network. Third,the intra skip connection mechanism is used to connect dierent layers inside the encoder as well asdecoder to further improve the model performance. Extensive experiments are performed to show theimproved performance of the proposed method as compared with three recent baseline methods.

Publication metadata

Author(s): Xian Y, Sun Y, Wang W, Naqvi SM

Publication type: Article

Publication status: Published

Journal: Neural Networks

Year: 2021

Volume: 143

Pages: 97-107

Print publication date: 01/11/2021

Online publication date: 25/05/2021

Acceptance date: 14/05/2021

Date deposited: 18/05/2021

ISSN (electronic): 0893-6080

Publisher: Elsevier

URL: https://doi.org/10.1016/j.neunet.2021.05.017

DOI: 10.1016/j.neunet.2021.05.017

Altmetrics

Share