A lightweight speech recognition method with target-swap knowledge distillation for Mandarin air traffic control communications
PeerJ Computer Science · 01.11.2023 · 9 · e1650
DOI: 10.7717/peerj-cs.1650 ↗Kurz gefasst
Fachartikel in der Zeitschrift PeerJ Computer Science, online erschienen am 01.11.2023. Verfasst von Jin Ren und 3 weiteren Autor:innen. Beteiligte Einrichtung: Institute of Applied Artificial Intelligence of the Guangdong-Hong Kong-Macao Greater Bay Area, Shenzhen Polytechnic University, Shenzhen, Guangdong,…. Die Arbeit wurde bisher 5-mal zitiert (Crossref) und nennt 38 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0). Gefördert von: Shenzhen Science and Technology Program, The General Higher Education Project of Guangdong Provincial Education Department, China Postdoctoral Science Foundation.
Abstract (Originalsprache)
Miscommunications between air traffic controllers (ATCOs) and pilots in air traffic control (ATC) may lead to catastrophic aviation accidents. Thanks to advances in speech and language processing, automatic speech recognition (ASR) is an appealing approach to prevent misunderstandings. To allow ATCOs and pilots sufficient time to respond instantly and effectively, the ASR systems for ATC must have both superior recognition performance and low transcription latency. However, most existing ASR works for ATC are primarily concerned with recognition performance while paying little attention to recognition speed, which motivates the research in this article. To address this issue, this article introduces knowledge distillation into the ASR for Mandarin ATC communications to enhance the generalization performance of the light model. Specifically, we propose a simple yet effective lightweight strategy, named Target-Swap Knowledge Distillation (TSKD), which swaps the logit output of the teacher and student models for the target class. It can mitigate the potential overconfidence of the teacher model regarding the target class and enable the student model to concentrate on the distillation of knowledge from non-target classes. Extensive experiments are conducted to demonstrate the effectiveness of the proposed TSKD in homogeneous and heterogeneous architectures. The experimental results reveal that the generated lightweight ASR model achieves a balance between recognition accuracy and transcription latency.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Institute of Applied Artificial Intelligence of the Guangdong-Hong Kong-Macao Greater Bay Area, Shenzhen Polytechnic University, Shenzhen, Guangdong, China.
- Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, Guangdong, China.
- Industrial Training Centre, Shenzhen Polytechnic University, Shenzhen, Guangdong, China.
Förderung
- Shenzhen Science and Technology Program
- The General Higher Education Project of Guangdong Provincial Education Department
- China Postdoctoral Science Foundation
- the Post-doctoral Foundation Project of Shenzhen Polytechnic
Zitieren
APA
Ren, J., Yang, S., Shi, Y., & Yang, J. (2023). A lightweight speech recognition method with target-swap knowledge distillation for Mandarin air traffic control communications. PeerJ Computer Science, 9, e1650. https://doi.org/10.7717/peerj-cs.1650
Vancouver
Ren J, Yang S, Shi Y, Yang J. A lightweight speech recognition method with target-swap knowledge distillation for Mandarin air traffic control communications. PeerJ Computer Science. 2023;9:e1650. doi:10.7717/peerj-cs.1650
BibTeX
@article{Ren2023,
author = {Ren, Jin and Yang, Shunzhi and Shi, Yihua and Yang, Jinfeng},
title = {A lightweight speech recognition method with target-swap knowledge distillation for Mandarin air traffic control communications},
journal = {PeerJ Computer Science},
year = {2023},
volume = {9},
pages = {e1650},
doi = {10.7717/peerj-cs.1650}
}
EUVIMED