Rainbow Keywords: Efficient Incremental Learning for Online Spoken Keyword Spotting

TL;DR

This paper introduces Rainbow Keywords, a novel incremental learning method for online keyword spotting that mitigates catastrophic forgetting and is suitable for edge devices with limited memory, achieving significant accuracy improvements.

Contribution

The paper proposes Rainbow Keywords, a diversity-aware incremental learning approach with data augmentation and knowledge distillation for efficient edge deployment.

Findings

Achieves 4.2% accuracy improvement over baseline.

Requires less memory for edge device deployment.

Effectively mitigates catastrophic forgetting.

Abstract

Catastrophic forgetting is a thorny challenge when updating keyword spotting (KWS) models after deployment. This problem will be more challenging if KWS models are further required for edge devices due to their limited memory. To alleviate such an issue, we propose a novel diversity-aware incremental learning method named Rainbow Keywords (RK). Specifically, the proposed RK approach introduces a diversity-aware sampler to select a diverse set from historical and incoming keywords by calculating classification uncertainty. As a result, the RK approach can incrementally learn new tasks without forgetting prior knowledge. Besides, the RK approach also proposes data augmentation and knowledge distillation loss function for efficient memory management on the edge device. Experimental results show that the proposed RK approach achieves 4.2% absolute improvement in terms of average accuracy over the best baseline on Google Speech Command dataset with less required memory. The scripts are available on GitHub.