Compact CNNs for End-to-End Keyword Spotting on Resource-Constrained Edge AI Devices

Joseph Lin; Ren Yuan Lyu

Compact CNNs for End-to-End Keyword Spotting on Resource-Constrained Edge AI Devices

Joseph Lin
, Ren Yuan Lyu

資訊工程學系(含學碩博士班)

Hsinchu County American School

研究成果: 圖書/報告稿件的類型 › 會議稿件 › 同行評審

1 引文斯高帕斯（Scopus）

摘要

In this paper, we explore compact convolutional neural networks (CNNs) for end-to-end keyword spotting from raw audio to final recognition results, without using traditional feature extraction based on spectrogram. Such fully CNN models reach 90.5% accuracy, an improvement of 12.15% over traditional methods with similar structures, which only achieve 78.35% accuracy, on the Speech Commands dataset. This shows that learned CNN features outperform predefined FFT-based transforms. The results show that compact end-toend CNNs enable efficient, accurate small vocabulary keyword spotting that is well-suited for resource-constrained edge devices.

原文	英語
主出版物標題	ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing
編輯	Jheng-Long Wu, Ming-Hsiang Su, Hen-Hsen Huang, Yu Tsao, Hou-Chiang Tseng, Chia-Hui Chang, Lung-Hao Lee, Yuan-Fu Liao, Wei-Yun Ma
發行者	The Association for Computational Linguistics and Chinese Language Processing (ACLCLP)
頁面	222-226
頁數	5
ISBN（電子）	9789869576963
出版狀態	已出版 - 2023
事件	35th Conference on Computational Linguistics and Speech Processing, ROCLING 2023 - Taipei City, 台灣持續時間: 20 10 2023 → 21 10 2023

出版系列

名字	ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing

Conference

Conference	35th Conference on Computational Linguistics and Speech Processing, ROCLING 2023
國家/地區	台灣
城市	Taipei City
期間	20/10/23 → 21/10/23

文獻附註

Publisher Copyright:
© 2023 ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing. All rights reserved.

其它文件與鏈接

Link to publication in Scopus

引用此

Lin, J., & Lyu, R. Y. (2023). Compact CNNs for End-to-End Keyword Spotting on Resource-Constrained Edge AI Devices. 於 J.-L. Wu, M.-H. Su, H.-H. Huang, Y. Tsao, H.-C. Tseng, C.-H. Chang, L.-H. Lee, Y.-F. Liao, & W.-Y. Ma (編輯), ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing (頁 222-226). (ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing). The Association for Computational Linguistics and Chinese Language Processing (ACLCLP).

Lin, Joseph ; Lyu, Ren Yuan. / Compact CNNs for End-to-End Keyword Spotting on Resource-Constrained Edge AI Devices. ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing. 編輯 / Jheng-Long Wu ; Ming-Hsiang Su ; Hen-Hsen Huang ; Yu Tsao ; Hou-Chiang Tseng ; Chia-Hui Chang ; Lung-Hao Lee ; Yuan-Fu Liao ; Wei-Yun Ma. The Association for Computational Linguistics and Chinese Language Processing (ACLCLP), 2023. 頁 222-226 (ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing).

@inproceedings{808bac27e8514ed9b3d895037a8c30a4,

title = "Compact CNNs for End-to-End Keyword Spotting on Resource-Constrained Edge AI Devices",

abstract = "In this paper, we explore compact convolutional neural networks (CNNs) for end-to-end keyword spotting from raw audio to final recognition results, without using traditional feature extraction based on spectrogram. Such fully CNN models reach 90.5\% accuracy, an improvement of 12.15\% over traditional methods with similar structures, which only achieve 78.35\% accuracy, on the Speech Commands dataset. This shows that learned CNN features outperform predefined FFT-based transforms. The results show that compact end-toend CNNs enable efficient, accurate small vocabulary keyword spotting that is well-suited for resource-constrained edge devices.",

keywords = "End-to-end models, keyword spotting, raw audio processing",

author = "Joseph Lin and Lyu, \{Ren Yuan\}",

note = "Publisher Copyright: {\textcopyright} 2023 ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing. All rights reserved.; 35th Conference on Computational Linguistics and Speech Processing, ROCLING 2023 ; Conference date: 20-10-2023 Through 21-10-2023",

year = "2023",

language = "英语",

series = "ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing",

publisher = "The Association for Computational Linguistics and Chinese Language Processing (ACLCLP)",

pages = "222--226",

editor = "Jheng-Long Wu and Ming-Hsiang Su and Hen-Hsen Huang and Yu Tsao and Hou-Chiang Tseng and Chia-Hui Chang and Lung-Hao Lee and Yuan-Fu Liao and Wei-Yun Ma",

booktitle = "ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing",

}

Lin, J & Lyu, RY 2023, Compact CNNs for End-to-End Keyword Spotting on Resource-Constrained Edge AI Devices. 於 J-L Wu, M-H Su, H-H Huang, Y Tsao, H-C Tseng, C-H Chang, L-H Lee, Y-F Liao & W-Y Ma (編輯), ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing. ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing, The Association for Computational Linguistics and Chinese Language Processing (ACLCLP), 頁 222-226, 35th Conference on Computational Linguistics and Speech Processing, ROCLING 2023, Taipei City, 台灣, 20/10/23.

Compact CNNs for End-to-End Keyword Spotting on Resource-Constrained Edge AI Devices. / Lin, Joseph; Lyu, Ren Yuan.
ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing. 編輯 / Jheng-Long Wu; Ming-Hsiang Su; Hen-Hsen Huang; Yu Tsao; Hou-Chiang Tseng; Chia-Hui Chang; Lung-Hao Lee; Yuan-Fu Liao; Wei-Yun Ma. The Association for Computational Linguistics and Chinese Language Processing (ACLCLP), 2023. p. 222-226 (ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing).

研究成果: 圖書/報告稿件的類型 › 會議稿件 › 同行評審

TY - GEN

T1 - Compact CNNs for End-to-End Keyword Spotting on Resource-Constrained Edge AI Devices

AU - Lin, Joseph

AU - Lyu, Ren Yuan

PY - 2023

Y1 - 2023

N2 - In this paper, we explore compact convolutional neural networks (CNNs) for end-to-end keyword spotting from raw audio to final recognition results, without using traditional feature extraction based on spectrogram. Such fully CNN models reach 90.5% accuracy, an improvement of 12.15% over traditional methods with similar structures, which only achieve 78.35% accuracy, on the Speech Commands dataset. This shows that learned CNN features outperform predefined FFT-based transforms. The results show that compact end-toend CNNs enable efficient, accurate small vocabulary keyword spotting that is well-suited for resource-constrained edge devices.

AB - In this paper, we explore compact convolutional neural networks (CNNs) for end-to-end keyword spotting from raw audio to final recognition results, without using traditional feature extraction based on spectrogram. Such fully CNN models reach 90.5% accuracy, an improvement of 12.15% over traditional methods with similar structures, which only achieve 78.35% accuracy, on the Speech Commands dataset. This shows that learned CNN features outperform predefined FFT-based transforms. The results show that compact end-toend CNNs enable efficient, accurate small vocabulary keyword spotting that is well-suited for resource-constrained edge devices.

KW - End-to-end models

KW - keyword spotting

KW - raw audio processing

UR - https://www.scopus.com/pages/publications/85184839745

M3 - 会议稿件

AN - SCOPUS:85184839745

T3 - ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing

SP - 222

EP - 226

BT - ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing

A2 - Wu, Jheng-Long

A2 - Su, Ming-Hsiang

A2 - Huang, Hen-Hsen

A2 - Tsao, Yu

A2 - Tseng, Hou-Chiang

A2 - Chang, Chia-Hui

A2 - Lee, Lung-Hao

A2 - Liao, Yuan-Fu

A2 - Ma, Wei-Yun

PB - The Association for Computational Linguistics and Chinese Language Processing (ACLCLP)

T2 - 35th Conference on Computational Linguistics and Speech Processing, ROCLING 2023

Y2 - 20 October 2023 through 21 October 2023

ER -

Lin J, Lyu RY. Compact CNNs for End-to-End Keyword Spotting on Resource-Constrained Edge AI Devices. 於 Wu JL, Su MH, Huang HH, Tsao Y, Tseng HC, Chang CH, Lee LH, Liao YF, Ma WY, 編輯, ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing. The Association for Computational Linguistics and Chinese Language Processing (ACLCLP). 2023. p. 222-226. (ROCLING 2023 - Proceedings of the 35th Conference on Computational Linguistics and Speech Processing).

Compact CNNs for End-to-End Keyword Spotting on Resource-Constrained Edge AI Devices

摘要

出版系列

Conference

文獻附註

其它文件與鏈接

指紋

引用此