arXiv cs.CL· Thai Quoc On, Homayoon Beigi·· 6 天前AI 评分18
金融语音需要持续学习正则化吗？EWC 与 mini-KFAC 在财报电话会议 ASR 上的研究
Does Financial Speech Need Continual Learning Regularization? A Study of EWC and mini-KFAC on Earnings Call ASR
AI 导读
研究将 Whisper-small 从通用语音语料适配到 Earnings22 基准，发现所有评估方法在同域金融语音留出集上均出现性能提升而非灾难性遗忘，任务互补性使正则化反而有害。当确需正则化时，逐层 mini-KFAC 近似比标准对角 EWC 取得更低 WER 和更稳定的优化。结果表明应用持续学习约束前应先评估任务互补性，必要时优先选择 mini-KFAC EWC。
正文 · 原文
Abstract:Earnings call transcription presents a realistic continual learning scenario where Automatic Speech Recognition (ASR) systems are updated with new recordings but risk catastrophic forgetting of previously learned financial speech patterns. Adapting Whisper-small from a general-domain speech corpus to the Earnings22 benchmark reveals the opposite: all evaluated methods exhibit improvement on held-out financial speech from the same domain during adaptation rather than forgetting. This task complementarity makes regularization counterproductive. Nevertheless, when regularization is warranted, a layer-wise mini-Kronecker-Factored Approximate Curvature (mini-KFAC) approximation seems to achieve lower word error rates (WERs) and more stable optimization than standard diagonal Elastic Weight Consolidation (EWC). The results suggest that task complementarity should be assessed before applying continual learning constraints, and prefer mini-KFAC EWC over diagonal EWC when needed.
| Comments: | 10 pages, 7 figures, 3 tables. Presented at the 9th International Conference on Natural Language and Speech Processing (ICNLSP 2026), University of Trento, Trento, Italy, September 25-26, 2026. Proceedings to appear in the ACL Anthology |
| Subjects: | Audio and Speech Processing (eess.AS); Computation and Language (cs.CL); Machine Learning (cs.LG) |
| ACM classes: | I.2.7 |
| Cite as: | arXiv:2610.13432 [eess.AS] |
| (or arXiv:2610.13432v1 [eess.AS] for this version) | |
| https://doi.org/10.48550/arXiv.2610.13432 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Homayoon Beigi [view email]
[v1]
Fri, 9 Oct 2026 11:34:13 UTC (1,103 KB)
来源：arXiv cs.CL · arxiv.org