Beyond Manual Transcripts: The Potential of Automated Speech Recognition Errors in Improving Alzheimer's Disease Detection

TL;DR

This study explores how errors in automated speech recognition can enhance Alzheimer's disease detection, revealing that certain ASR transcripts outperform manual ones and proposing a model for interpretability and pattern discovery.

Contribution

The paper demonstrates that ASR errors can serve as valuable cues for AD detection and introduces a cross-attention interpretability model to leverage these cues effectively.

Findings

ASR-synthesized speech outperforms manual transcripts in detection accuracy.

The proposed interpretability model identifies AD-related cues within speech.

ASR errors can provide beneficial information for improving AD detection.

Abstract

Recent breakthroughs in Automatic Speech Recognition (ASR) have enabled fully automated Alzheimer's Disease (AD) detection using ASR transcripts. Nonetheless, the impact of ASR errors on AD detection remains poorly understood. This paper fills the gap. We conduct a comprehensive study on AD detection using transcripts from various ASR models and their synthesized speech on the ADReSS dataset. Experimental results reveal that certain ASR transcripts (ASR-synthesized speech) outperform manual transcripts (manual-synthesized speech) in detection accuracy, suggesting that ASR errors may provide valuable cues for improving AD detection. Additionally, we propose a cross-attention-based interpretability model that not only identifies these cues but also achieves superior or comparable performance to the baseline. Furthermore, we utilize this model to unveil AD-related patterns within pre-trained embeddings. Our study offers novel insights into the potential of ASR models for AD detection.