Segment-Level Mandarin Chinese Speech-Based Cognitive Impairment Detection via an Autoencoder with Contrastive Learning

The article presents a segment-level representation learning framework for detecting cognitive impairment through Mandarin Chinese speech, utilizing an autoencoder and contrastive learning. By segmenting speech recordings into spectrograms and employing data augmentation techniques, the model achieved competitive performance on four independent datasets, particularly excelling in a challenging three-class classification task. This approach is significant for practitioners as it offers a scalable method for cognitive impairment screening in environments with limited labeled data.