Efficient Pause Extraction and Encode Strategy for Alzheimer's Disease Detection Using Only Acoustic Features from Spontaneous Speech

This paper proposes a speech pause feature extraction and encoding strategy for Alzheimer's Disease (AD) detection using only acoustic features. A voice activity detection (VAD) method is used to identify pauses, encoding them into binary sequences. An ensemble machine-learning approach, using VAD pause feature sequences and common acoustic feature sets (ComParE and eGeMAPS), classifies AD from spontaneous speech. Results on public datasets (ADReSS and ADReSSo) and a local dataset show that VAD Pause features outperform ComParE and eGeMAPS, with the ensemble method improving accuracy by over 5%. The method achieves 80% accuracy on the local Chinese dataset.

Jiamin Liu, Fan Fu, Liang Li, Junxiao Yu, Dacheng Zhong, Songsheng Zhu, Yuxuan Zhou, Bin Liu, Jianqing Li