LDCCRN: Robust Deep Learning-based Speech Enhancement

Chun Yin Yeung; Steve W.Y. Mung; Yat Sze Choy; Daniel P.K. Lun

doi:10.1117/12.2626108

LDCCRN: Robust Deep Learning-based Speech Enhancement

Chun Yin Yeung, Steve W.Y. Mung, Yat Sze Choy, Daniel P.K. Lun

Research output: Chapter in book / Conference proceeding › Conference article published in proceeding or book › Academic research › peer-review

1 Citation (Scopus)

Abstract

Deep learning-based speech enhancement methods make use of their non-linearity properties to estimate the speech and noise signals, especially the non-stationary noise. DCCRN, in particular, achieves state-of-the-art performance on speech intelligibility. However, the non-linear property also causes concern about the robustness of the method. Novel and unexpected noises can be generated if the noisy input speech is beyond the operation condition of the method. In this paper, we propose a hybrid framework called LDCCRN, which integrates a traditional speech enhancement method LogMMSE-EM and DCCRN. The proposed framework leverages the strength of both approaches to improve the robustness in speech enhancement. While the DCCRN continues to remove the non-stationary noise in the speech, the novel noises generated by DCCRN, if any, are effectively suppressed by LogMMSE-EM. As shown in our experimental results, the proposed method achieves better performance over the traditional approaches measured with standard evaluation methods.

Original language	English
Title of host publication	International Workshop on Advanced Imaging Technology, IWAIT 2022
Editors	Masayuki Nakajima, Shogo Muramatsu, Jae-Gon Kim, Jing-Ming Guo, Qian Kemao
Publisher	SPIE
ISBN (Electronic)	9781510653313
DOIs	https://doi.org/10.1117/12.2626108
Publication status	Published - 30 Apr 2022
Event	2022 International Workshop on Advanced Imaging Technology, IWAIT 2022 - Hong Kong, China Duration: 4 Jan 2022 → 6 Jan 2022

Publication series

Name	Proceedings of SPIE - The International Society for Optical Engineering
Volume	12177
ISSN (Print)	0277-786X
ISSN (Electronic)	1996-756X

Conference

Conference	2022 International Workshop on Advanced Imaging Technology, IWAIT 2022
Country/Territory	China
City	Hong Kong
Period	4/01/22 → 6/01/22

Keywords

complex-network
deep learning denoise
Speech enhancement

ASJC Scopus subject areas

Electronic, Optical and Magnetic Materials
Condensed Matter Physics
Computer Science Applications
Applied Mathematics
Electrical and Electronic Engineering

More information

10.1117/12.2626108

Cite this

Yeung, C. Y., Mung, S. W. Y., Choy, Y. S., & Lun, D. P. K. (2022). LDCCRN: Robust Deep Learning-based Speech Enhancement. In M. Nakajima, S. Muramatsu, J.-G. Kim, J.-M. Guo, & Q. Kemao (Eds.), International Workshop on Advanced Imaging Technology, IWAIT 2022 Article Article number 121772N (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 12177). SPIE. https://doi.org/10.1117/12.2626108

@inproceedings{4cf87149e02b484b9207e11cbb2f1e69,

title = "LDCCRN: Robust Deep Learning-based Speech Enhancement",

abstract = "Deep learning-based speech enhancement methods make use of their non-linearity properties to estimate the speech and noise signals, especially the non-stationary noise. DCCRN, in particular, achieves state-of-the-art performance on speech intelligibility. However, the non-linear property also causes concern about the robustness of the method. Novel and unexpected noises can be generated if the noisy input speech is beyond the operation condition of the method. In this paper, we propose a hybrid framework called LDCCRN, which integrates a traditional speech enhancement method LogMMSE-EM and DCCRN. The proposed framework leverages the strength of both approaches to improve the robustness in speech enhancement. While the DCCRN continues to remove the non-stationary noise in the speech, the novel noises generated by DCCRN, if any, are effectively suppressed by LogMMSE-EM. As shown in our experimental results, the proposed method achieves better performance over the traditional approaches measured with standard evaluation methods.",

keywords = "complex-network, deep learning denoise, Speech enhancement",

author = "Yeung, {Chun Yin} and Mung, {Steve W.Y.} and Choy, {Yat Sze} and Lun, {Daniel P.K.}",

note = "Funding Information: *The work described in this paper was fully supported by a grant from the Innovation and Technology Commission of the Hong Kong Special Administrative Region, China (Project No. UIM/381). Chun-Yin Yeung, Steve W.Y. Mung, and Daniel P.K. Lun are with the Dept. of EIE and Yat Sze Choy is with the Dept. of ME of the Hong Kong Polytechnic University, Hong Kong, China. (Corresponding author: Dr Daniel P.K. Lun, e-mail: [email protected]). Publisher Copyright: {\textcopyright} 2022 SPIE.; 2022 International Workshop on Advanced Imaging Technology, IWAIT 2022 ; Conference date: 04-01-2022 Through 06-01-2022",

year = "2022",

month = apr,

day = "30",

doi = "10.1117/12.2626108",

language = "English",

series = "Proceedings of SPIE - The International Society for Optical Engineering",

publisher = "SPIE",

editor = "Masayuki Nakajima and Shogo Muramatsu and Jae-Gon Kim and Jing-Ming Guo and Qian Kemao",

booktitle = "International Workshop on Advanced Imaging Technology, IWAIT 2022",

address = "United States",

}

Yeung, CY, Mung, SWY, Choy, YS & Lun, DPK 2022, LDCCRN: Robust Deep Learning-based Speech Enhancement. in M Nakajima, S Muramatsu, J-G Kim, J-M Guo & Q Kemao (eds), International Workshop on Advanced Imaging Technology, IWAIT 2022., Article number 121772N, Proceedings of SPIE - The International Society for Optical Engineering, vol. 12177, SPIE, 2022 International Workshop on Advanced Imaging Technology, IWAIT 2022, Hong Kong, China, 4/01/22. https://doi.org/10.1117/12.2626108

LDCCRN: Robust Deep Learning-based Speech Enhancement. / Yeung, Chun Yin; Mung, Steve W.Y.; Choy, Yat Sze et al.
International Workshop on Advanced Imaging Technology, IWAIT 2022. ed. / Masayuki Nakajima; Shogo Muramatsu; Jae-Gon Kim; Jing-Ming Guo; Qian Kemao. SPIE, 2022. Article number 121772N (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 12177).

Research output: Chapter in book / Conference proceeding › Conference article published in proceeding or book › Academic research › peer-review

TY - GEN

T1 - LDCCRN: Robust Deep Learning-based Speech Enhancement

AU - Yeung, Chun Yin

AU - Mung, Steve W.Y.

AU - Choy, Yat Sze

AU - Lun, Daniel P.K.

N1 - Funding Information: *The work described in this paper was fully supported by a grant from the Innovation and Technology Commission of the Hong Kong Special Administrative Region, China (Project No. UIM/381). Chun-Yin Yeung, Steve W.Y. Mung, and Daniel P.K. Lun are with the Dept. of EIE and Yat Sze Choy is with the Dept. of ME of the Hong Kong Polytechnic University, Hong Kong, China. (Corresponding author: Dr Daniel P.K. Lun, e-mail: [email protected]). Publisher Copyright: © 2022 SPIE.

PY - 2022/4/30

Y1 - 2022/4/30

N2 - Deep learning-based speech enhancement methods make use of their non-linearity properties to estimate the speech and noise signals, especially the non-stationary noise. DCCRN, in particular, achieves state-of-the-art performance on speech intelligibility. However, the non-linear property also causes concern about the robustness of the method. Novel and unexpected noises can be generated if the noisy input speech is beyond the operation condition of the method. In this paper, we propose a hybrid framework called LDCCRN, which integrates a traditional speech enhancement method LogMMSE-EM and DCCRN. The proposed framework leverages the strength of both approaches to improve the robustness in speech enhancement. While the DCCRN continues to remove the non-stationary noise in the speech, the novel noises generated by DCCRN, if any, are effectively suppressed by LogMMSE-EM. As shown in our experimental results, the proposed method achieves better performance over the traditional approaches measured with standard evaluation methods.

AB - Deep learning-based speech enhancement methods make use of their non-linearity properties to estimate the speech and noise signals, especially the non-stationary noise. DCCRN, in particular, achieves state-of-the-art performance on speech intelligibility. However, the non-linear property also causes concern about the robustness of the method. Novel and unexpected noises can be generated if the noisy input speech is beyond the operation condition of the method. In this paper, we propose a hybrid framework called LDCCRN, which integrates a traditional speech enhancement method LogMMSE-EM and DCCRN. The proposed framework leverages the strength of both approaches to improve the robustness in speech enhancement. While the DCCRN continues to remove the non-stationary noise in the speech, the novel noises generated by DCCRN, if any, are effectively suppressed by LogMMSE-EM. As shown in our experimental results, the proposed method achieves better performance over the traditional approaches measured with standard evaluation methods.

KW - complex-network

KW - deep learning denoise

KW - Speech enhancement

UR - http://www.scopus.com/inward/record.url?scp=85131817892&partnerID=8YFLogxK

U2 - 10.1117/12.2626108

DO - 10.1117/12.2626108

M3 - Conference article published in proceeding or book

AN - SCOPUS:85131817892

T3 - Proceedings of SPIE - The International Society for Optical Engineering

BT - International Workshop on Advanced Imaging Technology, IWAIT 2022

A2 - Nakajima, Masayuki

A2 - Muramatsu, Shogo

A2 - Kim, Jae-Gon

A2 - Guo, Jing-Ming

A2 - Kemao, Qian

PB - SPIE

T2 - 2022 International Workshop on Advanced Imaging Technology, IWAIT 2022

Y2 - 4 January 2022 through 6 January 2022

ER -

LDCCRN: Robust Deep Learning-based Speech Enhancement

Abstract

Publication series

Conference

Keywords

ASJC Scopus subject areas

More information

Other files and links

Fingerprint

Cite this