An Area-Efficient Ultra-Low-Power Time-Domain Feature Extractor for Edge Keyword Spotting

Qinyu Chen; Yaoxing Chang; Kwantae Kim; Chang Gao; Shih Chii Liu

doi:10.1109/ISCAS46773.2023.10181602

An Area-Efficient Ultra-Low-Power Time-Domain Feature Extractor for Edge Keyword Spotting

Qinyu Chen, Yaoxing Chang, Kwantae Kim, Chang Gao, Shih Chii Liu^*

^*Corresponding author for this work

Electronics

Research output: Chapter in Book/Conference proceedings/Edited volume › Conference contribution › Scientific › peer-review

30 Downloads (Pure)

Abstract

Keyword spotting (KWS) is an important task on edge low-power audio devices. A typical edge KWS system consists of a front-end feature extractor which outputs mel-scale frequency cepstral coefficients (MFCC) features followed by a back-end neural network classifier. KWS edge designs aim for the best power-performance-area metrics. This work proposes an area-efficient ultra-low-power time-domain infinite impulse response (IIR) filter-based feature extractor for a KWS system. It uses a serial architecture, and the architecture is further optimized for a low-cost computing structure and mixed-precision bit selection of the IIR coefficients while maintaining good KWS accuracy. Using a 65 nm process technology and a back-end neural network classifier, this simulated feature extractor has an area of 0.02 mm2 and achieves 3.3 μW @ 1.2 V, and achieves 92.5% accuracy on a 10-keyword, 12-class KWS task using the GSCD dataset.

Original language	English
Title of host publication	ISCAS 2023 - 56th IEEE International Symposium on Circuits and Systems, Proceedings
Publisher	IEEE
Number of pages	5
ISBN (Electronic)	9781665451093
DOIs	https://doi.org/10.1109/ISCAS46773.2023.10181602
Publication status	Published - 2023
Event	56th IEEE International Symposium on Circuits and Systems, ISCAS 2023 - Monterey, United States Duration: 21 May 2023 → 25 May 2023

Publication series

Name	Proceedings - IEEE International Symposium on Circuits and Systems
Volume	2023-May
ISSN (Print)	0271-4310

Conference

Conference	56th IEEE International Symposium on Circuits and Systems, ISCAS 2023
Country/Territory	United States
City	Monterey
Period	21/05/23 → 25/05/23

Bibliographical note

Green Open Access added to TU Delft Institutional Repository ‘You share, we take care!’ – Taverne project https://www.openaccess.nl/en/you-share-we-take-care
Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.

Keywords

hardware acceleration
infinite impulse response (IIR)
Keyword spotting (KWS)
long short-term memory

Access to Document

10.1109/ISCAS46773.2023.10181602

An_Area-Efficient_Ultra-Low-Power_Time-Domain_Feature_Extractor_for_Edge_Keyword_SpottingFinal published version, 1.29 MB

Cite this

Chen, Q., Chang, Y., Kim, K., Gao, C., & Liu, S. C. (2023). An Area-Efficient Ultra-Low-Power Time-Domain Feature Extractor for Edge Keyword Spotting. In ISCAS 2023 - 56th IEEE International Symposium on Circuits and Systems, Proceedings (Proceedings - IEEE International Symposium on Circuits and Systems; Vol. 2023-May). IEEE. https://doi.org/10.1109/ISCAS46773.2023.10181602

@inproceedings{580e5a49502845fabcf60747836ecf40,

title = "An Area-Efficient Ultra-Low-Power Time-Domain Feature Extractor for Edge Keyword Spotting",

abstract = "Keyword spotting (KWS) is an important task on edge low-power audio devices. A typical edge KWS system consists of a front-end feature extractor which outputs mel-scale frequency cepstral coefficients (MFCC) features followed by a back-end neural network classifier. KWS edge designs aim for the best power-performance-area metrics. This work proposes an area-efficient ultra-low-power time-domain infinite impulse response (IIR) filter-based feature extractor for a KWS system. It uses a serial architecture, and the architecture is further optimized for a low-cost computing structure and mixed-precision bit selection of the IIR coefficients while maintaining good KWS accuracy. Using a 65 nm process technology and a back-end neural network classifier, this simulated feature extractor has an area of 0.02 mm2 and achieves 3.3 μW @ 1.2 V, and achieves 92.5% accuracy on a 10-keyword, 12-class KWS task using the GSCD dataset.",

keywords = "hardware acceleration, infinite impulse response (IIR), Keyword spotting (KWS), long short-term memory",

author = "Qinyu Chen and Yaoxing Chang and Kwantae Kim and Chang Gao and Liu, {Shih Chii}",

note = "Green Open Access added to TU Delft Institutional Repository {\textquoteleft}You share, we take care!{\textquoteright} – Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public. ; 56th IEEE International Symposium on Circuits and Systems, ISCAS 2023 ; Conference date: 21-05-2023 Through 25-05-2023",

year = "2023",

doi = "10.1109/ISCAS46773.2023.10181602",

language = "English",

series = "Proceedings - IEEE International Symposium on Circuits and Systems",

publisher = "IEEE",

booktitle = "ISCAS 2023 - 56th IEEE International Symposium on Circuits and Systems, Proceedings",

address = "United States",

}

Chen, Q, Chang, Y, Kim, K, Gao, C & Liu, SC 2023, An Area-Efficient Ultra-Low-Power Time-Domain Feature Extractor for Edge Keyword Spotting. in ISCAS 2023 - 56th IEEE International Symposium on Circuits and Systems, Proceedings. Proceedings - IEEE International Symposium on Circuits and Systems, vol. 2023-May, IEEE, 56th IEEE International Symposium on Circuits and Systems, ISCAS 2023, Monterey, United States, 21/05/23. https://doi.org/10.1109/ISCAS46773.2023.10181602

An Area-Efficient Ultra-Low-Power Time-Domain Feature Extractor for Edge Keyword Spotting. / Chen, Qinyu; Chang, Yaoxing; Kim, Kwantae et al.
ISCAS 2023 - 56th IEEE International Symposium on Circuits and Systems, Proceedings. IEEE, 2023. (Proceedings - IEEE International Symposium on Circuits and Systems; Vol. 2023-May).

Research output: Chapter in Book/Conference proceedings/Edited volume › Conference contribution › Scientific › peer-review

TY - GEN

T1 - An Area-Efficient Ultra-Low-Power Time-Domain Feature Extractor for Edge Keyword Spotting

AU - Chen, Qinyu

AU - Chang, Yaoxing

AU - Kim, Kwantae

AU - Gao, Chang

AU - Liu, Shih Chii

N1 - Green Open Access added to TU Delft Institutional Repository ‘You share, we take care!’ – Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.

PY - 2023

Y1 - 2023

N2 - Keyword spotting (KWS) is an important task on edge low-power audio devices. A typical edge KWS system consists of a front-end feature extractor which outputs mel-scale frequency cepstral coefficients (MFCC) features followed by a back-end neural network classifier. KWS edge designs aim for the best power-performance-area metrics. This work proposes an area-efficient ultra-low-power time-domain infinite impulse response (IIR) filter-based feature extractor for a KWS system. It uses a serial architecture, and the architecture is further optimized for a low-cost computing structure and mixed-precision bit selection of the IIR coefficients while maintaining good KWS accuracy. Using a 65 nm process technology and a back-end neural network classifier, this simulated feature extractor has an area of 0.02 mm2 and achieves 3.3 μW @ 1.2 V, and achieves 92.5% accuracy on a 10-keyword, 12-class KWS task using the GSCD dataset.

AB - Keyword spotting (KWS) is an important task on edge low-power audio devices. A typical edge KWS system consists of a front-end feature extractor which outputs mel-scale frequency cepstral coefficients (MFCC) features followed by a back-end neural network classifier. KWS edge designs aim for the best power-performance-area metrics. This work proposes an area-efficient ultra-low-power time-domain infinite impulse response (IIR) filter-based feature extractor for a KWS system. It uses a serial architecture, and the architecture is further optimized for a low-cost computing structure and mixed-precision bit selection of the IIR coefficients while maintaining good KWS accuracy. Using a 65 nm process technology and a back-end neural network classifier, this simulated feature extractor has an area of 0.02 mm2 and achieves 3.3 μW @ 1.2 V, and achieves 92.5% accuracy on a 10-keyword, 12-class KWS task using the GSCD dataset.

KW - hardware acceleration

KW - infinite impulse response (IIR)

KW - Keyword spotting (KWS)

KW - long short-term memory

UR - http://www.scopus.com/inward/record.url?scp=85167663105&partnerID=8YFLogxK

U2 - 10.1109/ISCAS46773.2023.10181602

DO - 10.1109/ISCAS46773.2023.10181602

M3 - Conference contribution

AN - SCOPUS:85167663105

T3 - Proceedings - IEEE International Symposium on Circuits and Systems

BT - ISCAS 2023 - 56th IEEE International Symposium on Circuits and Systems, Proceedings

PB - IEEE

T2 - 56th IEEE International Symposium on Circuits and Systems, ISCAS 2023

Y2 - 21 May 2023 through 25 May 2023

ER -

An Area-Efficient Ultra-Low-Power Time-Domain Feature Extractor for Edge Keyword Spotting

Abstract

Publication series

Conference

Bibliographical note

Keywords

Access to Document

Other files and links

Fingerprint

Cite this