Leveraging AM and FM Rhythm Spectrograms for Dementia Classification and Assessment

TL;DR

This paper introduces Rhythm Formant Analysis (RFA) derived rhythm spectrograms as novel features for dementia classification and assessment, demonstrating improved accuracy over existing methods through handcrafted and fusion approaches.

Contribution

The study presents a new RFA-based feature extraction method and a fusion approach with vision transformers and BERT embeddings for enhanced dementia detection.

Findings

Handcrafted RFA features outperform eGeMAPs by 14.2% in accuracy.

Fusion of RFA spectrograms with ViT improves classification by 13.1%.

RFA spectrograms achieve comparable regression performance to baselines.

Abstract

This study explores the potential of Rhythm Formant Analysis (RFA) to capture long-term temporal modulations in dementia speech. Specifically, we introduce RFA-derived rhythm spectrograms as novel features for dementia classification and regression tasks. We propose two methodologies: (1) handcrafted features derived from rhythm spectrograms, and (2) a data-driven fusion approach, integrating proposed RFA-derived rhythm spectrograms with vision transformer (ViT) for acoustic representations along with BERT-based linguistic embeddings. We compare these with existing features. Notably, our handcrafted features outperform eGeMAPs with a relative improvement of $14.2\%$ in classification accuracy and comparable performance in the regression task. The fusion approach also shows improvement, with RFA spectrograms surpassing Mel spectrograms in classification by around a relative improvement of $13.1\%$ and a comparable regression score with the baselines.