Multi-Head Attention-Enhanced Speech Recognition for Reduced Data Requirements
Electronics · 31.12.2024 · 14 (1) · 128
DOI: 10.3390/electronics14010128 ↗Kurz gefasst
Fachartikel in der Zeitschrift Electronics, online erschienen am 31.12.2024. Verfasst von Yiqun Li und 5 weiteren Autor:innen. Die Arbeit wurde bisher 4-mal zitiert (Crossref) und nennt 38 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: China National Natural Science Foundation, Key Research and Development Project of Guilin, Project for Enhancing Young and Middle-aged Teacher’s Research Basis Ability in Colleges of Guangxi.
Abstract (Originalsprache)
Automatic speech recognition (ASR) technology has reached a mature level, and improving performance in data-scarce scenarios has become a key research focus. In this study, we propose a novel approach to building acoustic models by leveraging the multi-head attention mechanism to enhance various acoustic features extracted from audio data, combined with time-delay neural networks (TDNNs) using semi-orthogonal low-rank matrix factorization, resulting in significant performance improvements. To increase the robustness of the acoustic model, we introduce a small amount of data perturbation during the early stages of training. Additionally, during the decoding phase, we employ an external recurrent neural network language model (RNNLM) for rescoring, further enhancing the model’s accuracy. A comprehensive evaluation of the widely used LibriSpeech corpus shows that our method, using only half the training data, achieves word error rates (WERs) of 3.15 and 7.04 on the test_clean and test_other datasets, respectively. These results outperform models with similar architectures trained on the full dataset and demonstrate performance comparable to mainstream end-to-end models.
Förderung
- China National Natural Science Foundation
- Key Research and Development Project of Guilin
- Project for Enhancing Young and Middle-aged Teacher’s Research Basis Ability in Colleges of Guangxi
- Guangxi Key Research and Development Program
Zitieren
APA
Li, Y., Zhou, Y., Qiu, Z., Wang, Y., Wang, J., & Huang, G. (2024). Multi-Head Attention-Enhanced Speech Recognition for Reduced Data Requirements. Electronics, 14(1), 128. https://doi.org/10.3390/electronics14010128
Vancouver
Li Y, Zhou Y, Qiu Z, Wang Y, Wang J, Huang G. Multi-Head Attention-Enhanced Speech Recognition for Reduced Data Requirements. Electronics. 2024;14(1):128. doi:10.3390/electronics14010128
BibTeX
@article{Li2024,
author = {Li, Yiqun and Zhou, Ya and Qiu, Zan and Wang, Yabing and Wang, Jiahao and Huang, Guimin},
title = {Multi-Head Attention-Enhanced Speech Recognition for Reduced Data Requirements},
journal = {Electronics},
year = {2024},
volume = {14},
number = {1},
pages = {128},
doi = {10.3390/electronics14010128}
}
EUVIMED