A Lightweight Dynamic Filter For Keyword Spotting

Donghyeon Kim; Kyungdeuk Ko; Jeonggi Kwak; David K. Han; Hanseok Ko

doi:10.1109/ICASSPW59220.2023.10193258

A Lightweight Dynamic Filter For Keyword Spotting

Donghyeon Kim, Kyungdeuk Ko, Jeonggi Kwak, David K. Han, Hanseok Ko

School of Electrical Engineering

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

1 Citation (Scopus)

Abstract

Keyword Spotting (KWS) from speech signals is widely applied to perform fully hands-free speech recognition. The KWS network is designed as a small-footprint model so it can continuously be active. Recent efforts have explored dynamic filter-based models in deep learning frameworks to enhance the system's robustness or accuracy. However, as a dynamic filter framework requires high computational costs, the implementation is limited to the computational condition of the device. In this paper, we propose a lightweight dynamic filter to improve the performance of KWS. Our proposed model divides the dynamic filter into two branches to reduce computational complexity: pixel level and instance level. The proposed lightweight dynamic filter is applied to the front end of KWS to enhance the separability of the input data. The experimental results show that our model is robustly working on unseen noise and small training data environments by using a small computational resource.

Original language	English
Title of host publication	ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings
Publisher	Institute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)	9798350302615
DOIs	https://doi.org/10.1109/ICASSPW59220.2023.10193258
Publication status	Published - 2023
Event	2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, ICASSPW 2023 - Rhodes Island, Greece Duration: 2023 Jun 4 → 2023 Jun 10

Publication series

Name	ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings

Conference

Conference	2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, ICASSPW 2023
Country/Territory	Greece
City	Rhodes Island
Period	23/6/4 → 23/6/10

Bibliographical note

Funding Information:
This work was supported by Korea Environment Industry & Technology Institute(KEITI) through Exotic Invasive Species Management Program, funded by Korea Ministry of Environment(MOE) (2021002280004) David Han’s work on this paper is partly sponsored by the Office of Naval Research (.Grant Number: N00014-21-1-2790.).

Publisher Copyright:
© 2023 IEEE.

Keywords

computational cost
dynamic filter
dynamic weight
keyword spotting

ASJC Scopus subject areas

Computer Science Applications
Acoustics and Ultrasonics
Computer Networks and Communications
Information Systems
Signal Processing

Access to Document

10.1109/ICASSPW59220.2023.10193258

Cite this

Kim, D., Ko, K., Kwak, J., Han, D. K., & Ko, H. (2023). A Lightweight Dynamic Filter For Keyword Spotting. In ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings (ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ICASSPW59220.2023.10193258

A Lightweight Dynamic Filter For Keyword Spotting. / Kim, Donghyeon; Ko, Kyungdeuk; Kwak, Jeonggi et al.
ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings. Institute of Electrical and Electronics Engineers Inc., 2023. (ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Kim, D, Ko, K, Kwak, J, Han, DK & Ko, H 2023, A Lightweight Dynamic Filter For Keyword Spotting. in ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings. ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings, Institute of Electrical and Electronics Engineers Inc., 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, ICASSPW 2023, Rhodes Island, Greece, 23/6/4. https://doi.org/10.1109/ICASSPW59220.2023.10193258

Kim D, Ko K, Kwak J, Han DK, Ko H. A Lightweight Dynamic Filter For Keyword Spotting. In ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings. Institute of Electrical and Electronics Engineers Inc. 2023. (ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings). doi: 10.1109/ICASSPW59220.2023.10193258

Kim, Donghyeon ; Ko, Kyungdeuk ; Kwak, Jeonggi et al. / A Lightweight Dynamic Filter For Keyword Spotting. ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings. Institute of Electrical and Electronics Engineers Inc., 2023. (ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings).

@inproceedings{cb187721249249b982562e3c1a6a8292,

title = "A Lightweight Dynamic Filter For Keyword Spotting",

abstract = "Keyword Spotting (KWS) from speech signals is widely applied to perform fully hands-free speech recognition. The KWS network is designed as a small-footprint model so it can continuously be active. Recent efforts have explored dynamic filter-based models in deep learning frameworks to enhance the system's robustness or accuracy. However, as a dynamic filter framework requires high computational costs, the implementation is limited to the computational condition of the device. In this paper, we propose a lightweight dynamic filter to improve the performance of KWS. Our proposed model divides the dynamic filter into two branches to reduce computational complexity: pixel level and instance level. The proposed lightweight dynamic filter is applied to the front end of KWS to enhance the separability of the input data. The experimental results show that our model is robustly working on unseen noise and small training data environments by using a small computational resource.",

keywords = "computational cost, dynamic filter, dynamic weight, keyword spotting",

author = "Donghyeon Kim and Kyungdeuk Ko and Jeonggi Kwak and Han, {David K.} and Hanseok Ko",

note = "Funding Information: This work was supported by Korea Environment Industry & Technology Institute(KEITI) through Exotic Invasive Species Management Program, funded by Korea Ministry of Environment(MOE) (2021002280004) David Han{\textquoteright}s work on this paper is partly sponsored by the Office of Naval Research (.Grant Number: N00014-21-1-2790.). Publisher Copyright: {\textcopyright} 2023 IEEE.; 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, ICASSPW 2023 ; Conference date: 04-06-2023 Through 10-06-2023",

year = "2023",

doi = "10.1109/ICASSPW59220.2023.10193258",

language = "English",

series = "ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings",

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

booktitle = "ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings",

}

TY - GEN

T1 - A Lightweight Dynamic Filter For Keyword Spotting

AU - Kim, Donghyeon

AU - Ko, Kyungdeuk

AU - Kwak, Jeonggi

AU - Han, David K.

AU - Ko, Hanseok

N1 - Funding Information: This work was supported by Korea Environment Industry & Technology Institute(KEITI) through Exotic Invasive Species Management Program, funded by Korea Ministry of Environment(MOE) (2021002280004) David Han’s work on this paper is partly sponsored by the Office of Naval Research (.Grant Number: N00014-21-1-2790.). Publisher Copyright: © 2023 IEEE.

PY - 2023

Y1 - 2023

N2 - Keyword Spotting (KWS) from speech signals is widely applied to perform fully hands-free speech recognition. The KWS network is designed as a small-footprint model so it can continuously be active. Recent efforts have explored dynamic filter-based models in deep learning frameworks to enhance the system's robustness or accuracy. However, as a dynamic filter framework requires high computational costs, the implementation is limited to the computational condition of the device. In this paper, we propose a lightweight dynamic filter to improve the performance of KWS. Our proposed model divides the dynamic filter into two branches to reduce computational complexity: pixel level and instance level. The proposed lightweight dynamic filter is applied to the front end of KWS to enhance the separability of the input data. The experimental results show that our model is robustly working on unseen noise and small training data environments by using a small computational resource.

AB - Keyword Spotting (KWS) from speech signals is widely applied to perform fully hands-free speech recognition. The KWS network is designed as a small-footprint model so it can continuously be active. Recent efforts have explored dynamic filter-based models in deep learning frameworks to enhance the system's robustness or accuracy. However, as a dynamic filter framework requires high computational costs, the implementation is limited to the computational condition of the device. In this paper, we propose a lightweight dynamic filter to improve the performance of KWS. Our proposed model divides the dynamic filter into two branches to reduce computational complexity: pixel level and instance level. The proposed lightweight dynamic filter is applied to the front end of KWS to enhance the separability of the input data. The experimental results show that our model is robustly working on unseen noise and small training data environments by using a small computational resource.

KW - computational cost

KW - dynamic filter

KW - dynamic weight

KW - keyword spotting

UR - http://www.scopus.com/inward/record.url?scp=85168255166&partnerID=8YFLogxK

U2 - 10.1109/ICASSPW59220.2023.10193258

DO - 10.1109/ICASSPW59220.2023.10193258

M3 - Conference contribution

AN - SCOPUS:85168255166

T3 - ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings

BT - ICASSPW 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, Proceedings

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing Workshops, ICASSPW 2023

Y2 - 4 June 2023 through 10 June 2023

ER -

A Lightweight Dynamic Filter For Keyword Spotting

Abstract

Publication series

Conference

Bibliographical note

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this