Continuous Learning for Children's ASR: Overcoming Catastrophic Forgetting with Elastic Weight Consolidation and Synaptic Intelligence

TL;DR

This paper introduces methods for online learning in children's speech recognition to prevent catastrophic forgetting, demonstrating improved accuracy using elastic weight consolidation and synaptic intelligence on a custom dataset.

Contribution

First study applying EWC and SI techniques to online children's ASR, addressing catastrophic forgetting in a privacy-preserving, sequential learning setting.

Findings

EWC reduces WER by 5.21% relative to baseline

SI reduces WER by 4.36% relative to baseline

Demonstrates effective continual learning for children's speech recognition

Abstract

In this work, we present the first study addressing automatic speech recognition (ASR) for children in an online learning setting. This is particularly important for both child-centric applications and the privacy protection of minors, where training models with sequentially arriving data is critical. The conventional approach of model fine-tuning often suffers from catastrophic forgetting. To tackle this issue, we explore two established techniques: elastic weight consolidation (EWC) and synaptic intelligence (SI). Using a custom protocol on the MyST corpus, tailored to the online learning setting, we achieve relative word error rate (WER) reductions of 5.21% with EWC and 4.36% with SI, compared to the fine-tuning baseline.