DFBNet: Deep Neural Network based Fixed Beamformer for Multi-channel Speech Separation

Ruqiao LIU; Yi ZHOU; Hongqing LIU; Xinmeng XU; Jie JIA; Binbin CHEN

doi:10.1109/SiPS52927.2021.00042

DFBNet: Deep Neural Network based Fixed Beamformer for Multi-channel Speech Separation

Ruqiao LIU
, Yi ZHOU
, Hongqing LIU
, Xinmeng XU
, Jie JIA
, Binbin CHEN

Research output: Book Chapters | Papers in Conference Proceedings › Conference paper (refereed) › Referred Conference Paper › peer-review

Abstract

The deep neural networks (DNNs) based beamformers have achieved significant improvements in speech separation tasks. This paper proposes a novel deep neural network (DNN) based fixed beamformer (DFBNet) that uniformly samples the space as a learning module. In addition, the initial coefficients of fixed beamformers in DFBNet are determined by the existing superdirective beamformer. Furthermore, to obtain the beams that related to each speaker, the proposed model has introduced a speech source estimation model, dual-path RNN (DPRNN), and an attention mechanism. The experimental results show that in the separation task with reverberation, the proposed way has better performance on scale-invariant signal-to-noise ratio (SI-SNR) and perceptual evaluation of speech quality (PESQ) than DPRNN and filter-and-sum network (FasNet) which is currently the most state-of-the-art temporal neural beamformer.

Original language	English
Title of host publication	2021 IEEE Workshop on Signal Processing Systems, SiPS 2021: Proceedings
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	194-198
Number of pages	5
ISBN (Electronic)	9781665401449
DOIs	https://doi.org/10.1109/SiPS52927.2021.00042
Publication status	Published - 2021
Externally published	Yes
Event	2021 IEEE Workshop on Signal Processing Systems - Coimbra, Portugal Duration: 19 Oct 2021 → 21 Oct 2021

Publication series

Name	IEEE Workshop on Signal Processing Systems
Publisher	IEEE
ISSN (Print)	1520-6130
ISSN (Electronic)	2374-7390

Workshop

Workshop	2021 IEEE Workshop on Signal Processing Systems
Abbreviated title	SiPS 2021
Country/Territory	Portugal
City	Coimbra
Period	19/10/21 → 21/10/21

Bibliographical note

Publisher Copyright:
© 2021 IEEE.

Keywords

Deep neural network
Fixed beamformer
Multi-channel speech separation

Access to Document

10.1109/SiPS52927.2021.00042

Cite this

LIU, R., ZHOU, Y., LIU, H., XU, X., JIA, J., & CHEN, B. (2021). DFBNet: Deep Neural Network based Fixed Beamformer for Multi-channel Speech Separation. In 2021 IEEE Workshop on Signal Processing Systems, SiPS 2021: Proceedings (pp. 194-198). (IEEE Workshop on Signal Processing Systems). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/SiPS52927.2021.00042

@inproceedings{ce416c4547214d0f977a8c8552260b9d,

title = "DFBNet: Deep Neural Network based Fixed Beamformer for Multi-channel Speech Separation",

abstract = "The deep neural networks (DNNs) based beamformers have achieved significant improvements in speech separation tasks. This paper proposes a novel deep neural network (DNN) based fixed beamformer (DFBNet) that uniformly samples the space as a learning module. In addition, the initial coefficients of fixed beamformers in DFBNet are determined by the existing superdirective beamformer. Furthermore, to obtain the beams that related to each speaker, the proposed model has introduced a speech source estimation model, dual-path RNN (DPRNN), and an attention mechanism. The experimental results show that in the separation task with reverberation, the proposed way has better performance on scale-invariant signal-to-noise ratio (SI-SNR) and perceptual evaluation of speech quality (PESQ) than DPRNN and filter-and-sum network (FasNet) which is currently the most state-of-the-art temporal neural beamformer.",

keywords = "Deep neural network, Fixed beamformer, Multi-channel speech separation",

author = "Ruqiao LIU and Yi ZHOU and Hongqing LIU and Xinmeng XU and Jie JIA and Binbin CHEN",

note = "Publisher Copyright: {\textcopyright} 2021 IEEE.; 2021 IEEE Workshop on Signal Processing Systems, SiPS 2021 ; Conference date: 19-10-2021 Through 21-10-2021",

year = "2021",

doi = "10.1109/SiPS52927.2021.00042",

language = "English",

series = "IEEE Workshop on Signal Processing Systems",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "194--198",

booktitle = "2021 IEEE Workshop on Signal Processing Systems, SiPS 2021: Proceedings",

address = "United States",

}

LIU, R, ZHOU, Y, LIU, H, XU, X, JIA, J & CHEN, B 2021, DFBNet: Deep Neural Network based Fixed Beamformer for Multi-channel Speech Separation. in 2021 IEEE Workshop on Signal Processing Systems, SiPS 2021: Proceedings. IEEE Workshop on Signal Processing Systems, Institute of Electrical and Electronics Engineers Inc., pp. 194-198, 2021 IEEE Workshop on Signal Processing Systems, Coimbra, Portugal, 19/10/21. https://doi.org/10.1109/SiPS52927.2021.00042

DFBNet: Deep Neural Network based Fixed Beamformer for Multi-channel Speech Separation. / LIU, Ruqiao; ZHOU, Yi; LIU, Hongqing et al.
2021 IEEE Workshop on Signal Processing Systems, SiPS 2021: Proceedings. Institute of Electrical and Electronics Engineers Inc., 2021. p. 194-198 (IEEE Workshop on Signal Processing Systems).

Research output: Book Chapters | Papers in Conference Proceedings › Conference paper (refereed) › Referred Conference Paper › peer-review

TY - GEN

T1 - DFBNet: Deep Neural Network based Fixed Beamformer for Multi-channel Speech Separation

AU - LIU, Ruqiao

AU - ZHOU, Yi

AU - LIU, Hongqing

AU - XU, Xinmeng

AU - JIA, Jie

AU - CHEN, Binbin

PY - 2021

Y1 - 2021

N2 - The deep neural networks (DNNs) based beamformers have achieved significant improvements in speech separation tasks. This paper proposes a novel deep neural network (DNN) based fixed beamformer (DFBNet) that uniformly samples the space as a learning module. In addition, the initial coefficients of fixed beamformers in DFBNet are determined by the existing superdirective beamformer. Furthermore, to obtain the beams that related to each speaker, the proposed model has introduced a speech source estimation model, dual-path RNN (DPRNN), and an attention mechanism. The experimental results show that in the separation task with reverberation, the proposed way has better performance on scale-invariant signal-to-noise ratio (SI-SNR) and perceptual evaluation of speech quality (PESQ) than DPRNN and filter-and-sum network (FasNet) which is currently the most state-of-the-art temporal neural beamformer.

AB - The deep neural networks (DNNs) based beamformers have achieved significant improvements in speech separation tasks. This paper proposes a novel deep neural network (DNN) based fixed beamformer (DFBNet) that uniformly samples the space as a learning module. In addition, the initial coefficients of fixed beamformers in DFBNet are determined by the existing superdirective beamformer. Furthermore, to obtain the beams that related to each speaker, the proposed model has introduced a speech source estimation model, dual-path RNN (DPRNN), and an attention mechanism. The experimental results show that in the separation task with reverberation, the proposed way has better performance on scale-invariant signal-to-noise ratio (SI-SNR) and perceptual evaluation of speech quality (PESQ) than DPRNN and filter-and-sum network (FasNet) which is currently the most state-of-the-art temporal neural beamformer.

KW - Deep neural network

KW - Fixed beamformer

KW - Multi-channel speech separation

UR - https://www.scopus.com/pages/publications/85122854871

U2 - 10.1109/SiPS52927.2021.00042

DO - 10.1109/SiPS52927.2021.00042

M3 - Conference paper (refereed)

AN - SCOPUS:85122854871

T3 - IEEE Workshop on Signal Processing Systems

SP - 194

EP - 198

BT - 2021 IEEE Workshop on Signal Processing Systems, SiPS 2021: Proceedings

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2021 IEEE Workshop on Signal Processing Systems

Y2 - 19 October 2021 through 21 October 2021

ER -

LIU R, ZHOU Y, LIU H, XU X, JIA J, CHEN B. DFBNet: Deep Neural Network based Fixed Beamformer for Multi-channel Speech Separation. In 2021 IEEE Workshop on Signal Processing Systems, SiPS 2021: Proceedings. Institute of Electrical and Electronics Engineers Inc. 2021. p. 194-198. (IEEE Workshop on Signal Processing Systems). Epub 2021 Nov 11. doi: 10.1109/SiPS52927.2021.00042

DFBNet: Deep Neural Network based Fixed Beamformer for Multi-channel Speech Separation

Abstract

Publication series

Workshop

Bibliographical note

Keywords

Access to Document

Other files and links

Fingerprint

Cite this