Close, G.L., Ravenscroft, W., Hain, T. orcid.org/0000-0003-0939-3464 et al. (1 more author) (2023) The University of Sheffield CHiME-7 UDASE challenge speech enhancement system. In: Proc. 7th International Workshop on Speech Processing in Everyday Environments (CHiME 2023). 7th International Workshop on Speech Processing in Everyday Environments (CHiME 2023), 25 Aug 2023, Dublin, Ireland. International Speech Communication Association (ISCA) , pp. 33-38.
Abstract
The CHiME-7 unsupervised domain adaptation speech enhancement (UDASE) challenge targets domain adaptation to unlabelled speech data. This paper describes the University of Sheffield team’s system submitted to the challenge. A generative adversarial network (GAN) methodology based on a conformer-based metric GAN (CMGAN) is employed as opposed to the unsupervised RemixIT strategy used in the CHiME-7 baseline system. The discriminator of the GAN is trained to predict the output score of a Deep Noise Suppression Mean Opinion Score (DNSMOS) metric. Additional data augmentation strategies are employed which provide the discriminator with historical training data outputs as well as more diverse training examples from an additional pseudo-generator. The proposed approach, denoted as CMGAN+/+, achieves significant improvement in DNSMOS evaluation metrics with the best proposed system achieving 3.51 OVR-MOS, a 24% improvement over the baseline.
Metadata
Item Type: | Proceedings Paper |
---|---|
Authors/Creators: |
|
Copyright, Publisher and Additional Information: | © 2023 ISCA. Reproduced in accordance with the publisher's self-archiving policy. |
Dates: |
|
Institution: | The University of Sheffield |
Academic Units: | The University of Sheffield > Faculty of Engineering (Sheffield) > Department of Computer Science (Sheffield) |
Funding Information: | Funder Grant number Engineering and Physical Sciences Research Council 2268977 |
Depositing User: | Symplectic Sheffield |
Date Deposited: | 17 Jan 2024 09:41 |
Last Modified: | 17 Jan 2024 09:42 |
Status: | Published |
Publisher: | International Speech Communication Association (ISCA) |
Refereed: | Yes |
Identification Number: | 10.21437/chime.2023-7 |
Related URLs: | |
Open Archives Initiative ID (OAI ID): | oai:eprints.whiterose.ac.uk:207514 |