Towards few-shot isolated word reading assessment

2025-07-16

AI Summary

Assessing isolated-word reading proficiency in children under low-resource conditions remains challenging—particularly when child speech data is scarce and automatic speech recognition (ASR) systems are impractical to deploy. Method: We propose an ASR-free few-shot evaluation framework leveraging intermediate representations from large-scale self-supervised learning (SSL) models. Our approach introduces discretized SSL features, template centroid averaging, and integrates dynamic time warping with prototypical network principles to score children’s utterances by comparing them against a small set of adult or child reference templates. Contribution/Results: Experiments on the Afrikaans child speech dataset reveal robust performance on adult speech but significant degradation on child speech—even with child-specific templates. This work provides the first systematic evidence of inherent limitations in current SSL representations for few-shot modeling of child speech, establishing a new perspective and benchmark analysis for low-resource pediatric speech assessment.

Technology Category

• Natural Language Processing: Speech
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Large pretrained models with web data
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

We explore an ASR-free method for isolated word reading assessment in low-resource settings. Our few-shot approach compares input child speech to a small set of adult-provided reference templates. Inputs and templates are encoded using intermediate layers from large self-supervised learned (SSL) models. Using an Afrikaans child speech benchmark, we investigate design options such as discretising SSL features and barycentre averaging of the templates. Idealised experiments show reasonable performance for adults, but a substantial drop for child speech input, even with child templates. Despite the success of employing SSL representations in low-resource speech tasks, our work highlights the limitations of SSL representations for processing child data when used in a few-shot classification system.

Problem

• Assess isolated word reading in low-resource settings
• Compare child speech to adult reference templates
• Evaluate SSL model limitations for child speech

Innovation

• ASR-free method for word assessment
• Few-shot approach with reference templates
• SSL model layers for encoding