Spot keywords from very noisy and mixed speech

Ying Shi; Dong Wang; Lantian Li; Jiqing Han; Shi Yin

doi:10.21437/Interspeech.2023-1242

Spot keywords from very noisy and mixed speech

Ying Shi
, Dong Wang^*
, Lantian Li
, Jiqing Han^*
, Shi Yin

^*Corresponding author for this work

School of Computer Science and Technology, Harbin Institute of Technology
Tsinghua University
Beijing University of Posts and Telecommunications
Huawei Technologies Canada Company Ltd.

Research output: Contribution to journal › Conference article › peer-review

3 Scopus citations

Abstract

Most existing keyword spotting research focuses on conditions with slight or moderate noise. In this paper, we try to tackle a more challenging task: detecting keywords buried under strong interfering speech (10 times higher than the keyword in amplitude), and even worse, mixed with other keywords. We propose a novel Mix Training (MT) strategy that encourages the model to discover low-energy keywords from noisy and mixed speech. Experiments were conducted with a vanilla CNN and two EfficientNet (B0/B2) architectures. The results evaluated with the Google Speech Command dataset demonstrated that the proposed mix training approach is highly effective and outperforms standard data augmentation and mixup training.

Original language	English
Pages (from-to)	1488-1492
Number of pages	5
Journal	Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH
Volume	2023-August
DOIs	https://doi.org/10.21437/Interspeech.2023-1242
State	Published - 2023
Externally published	Yes
Event	24th Annual conference of the International Speech Communication Association, Interspeech 2023 - Dublin, Ireland Duration: 20 Aug 2023 → 24 Aug 2023

Keywords

Keyword spotting
mix training
mixed speech

Access to Document

10.21437/Interspeech.2023-1242

Cite this

@article{13b536629f0442aaa971170c7a8afb35,

title = "Spot keywords from very noisy and mixed speech",

abstract = "Most existing keyword spotting research focuses on conditions with slight or moderate noise. In this paper, we try to tackle a more challenging task: detecting keywords buried under strong interfering speech (10 times higher than the keyword in amplitude), and even worse, mixed with other keywords. We propose a novel Mix Training (MT) strategy that encourages the model to discover low-energy keywords from noisy and mixed speech. Experiments were conducted with a vanilla CNN and two EfficientNet (B0/B2) architectures. The results evaluated with the Google Speech Command dataset demonstrated that the proposed mix training approach is highly effective and outperforms standard data augmentation and mixup training.",

keywords = "Keyword spotting, mix training, mixed speech",

author = "Ying Shi and Dong Wang and Lantian Li and Jiqing Han and Shi Yin",

note = "Publisher Copyright: {\textcopyright} 2023 International Speech Communication Association. All rights reserved.; 24th Annual conference of the International Speech Communication Association, Interspeech 2023 ; Conference date: 20-08-2023 Through 24-08-2023",

year = "2023",

doi = "10.21437/Interspeech.2023-1242",

language = "英语",

volume = "2023-August",

pages = "1488--1492",

journal = "Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH",

issn = "2308-457X",

publisher = "International Speech Communication Association",

}

TY - JOUR

T1 - Spot keywords from very noisy and mixed speech

AU - Shi, Ying

AU - Wang, Dong

AU - Li, Lantian

AU - Han, Jiqing

AU - Yin, Shi

PY - 2023

Y1 - 2023

N2 - Most existing keyword spotting research focuses on conditions with slight or moderate noise. In this paper, we try to tackle a more challenging task: detecting keywords buried under strong interfering speech (10 times higher than the keyword in amplitude), and even worse, mixed with other keywords. We propose a novel Mix Training (MT) strategy that encourages the model to discover low-energy keywords from noisy and mixed speech. Experiments were conducted with a vanilla CNN and two EfficientNet (B0/B2) architectures. The results evaluated with the Google Speech Command dataset demonstrated that the proposed mix training approach is highly effective and outperforms standard data augmentation and mixup training.

AB - Most existing keyword spotting research focuses on conditions with slight or moderate noise. In this paper, we try to tackle a more challenging task: detecting keywords buried under strong interfering speech (10 times higher than the keyword in amplitude), and even worse, mixed with other keywords. We propose a novel Mix Training (MT) strategy that encourages the model to discover low-energy keywords from noisy and mixed speech. Experiments were conducted with a vanilla CNN and two EfficientNet (B0/B2) architectures. The results evaluated with the Google Speech Command dataset demonstrated that the proposed mix training approach is highly effective and outperforms standard data augmentation and mixup training.

KW - Keyword spotting

KW - mix training

KW - mixed speech

UR - https://www.scopus.com/pages/publications/85171551277

U2 - 10.21437/Interspeech.2023-1242

DO - 10.21437/Interspeech.2023-1242

M3 - 会议文章

AN - SCOPUS:85171551277

SN - 2308-457X

VL - 2023-August

SP - 1488

EP - 1492

JO - Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH

JF - Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH

T2 - 24th Annual conference of the International Speech Communication Association, Interspeech 2023

Y2 - 20 August 2023 through 24 August 2023

ER -

Spot keywords from very noisy and mixed speech

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this