A Novel Data Augmentation Approach for Automatic Speaking Assessment on Opinion Expressions

TL;DR

This paper introduces a new data augmentation method using large language models and speech synthesis to improve automated speaking assessment of opinion expressions, especially in low-resource scenarios.

Contribution

It presents a novel training paradigm combining LLM-generated responses, speaker-aware speech synthesis, and a dynamic importance loss for enhanced ASA performance.

Findings

Outperforms existing methods on the LTTC dataset.

Effectively mitigates low-resource constraints.

Enables cross-modal proficiency scoring.

Abstract

Automated speaking assessment (ASA) on opinion expressions is often hampered by the scarcity of labeled recordings, which restricts prompt diversity and undermines scoring reliability. To address this challenge, we propose a novel training paradigm that leverages a large language models (LLM) to generate diverse responses of a given proficiency level, converts responses into synthesized speech via speaker-aware text-to-speech synthesis, and employs a dynamic importance loss to adaptively reweight training instances based on feature distribution differences between synthesized and real speech. Subsequently, a multimodal large language model integrates aligned textual features with speech signals to predict proficiency scores directly. Experiments conducted on the LTTC dataset show that our approach outperforms methods relying on real data or conventional augmentation, effectively mitigating low-resource constraints and enabling ASA on opinion expressions with cross-modal information.