Transcript-Free Lightweight Detection of Alzheimer’s Disease from Spontaneous Speech Using Handcrafted MFCC-Dominant Acoustic Biomarkers

This paper presents a transcript-free, audio-only baseline for Alzheimer’s disease detection using 176 Cookie Theft recordings from the DementiaBank Pitt corpus, comprising 88 AD and 88 control samples. WebRTC voice activity detection extracts speech regions, while 99 handcrafted acoustic-temporal features cover pauses, fluency, spectral and prosodic descriptors, and MFCC statistics with delta and delta-delta terms. A speaker-independent GroupShuffleSplit evaluation over 30 iterations yields a mean AUC of 0.674 with an RBF-kernel SVM. A non-nested exploratory Top-20 feature analysis reaches an AUC of 0.719 and is explicitly excluded from the primary conclusion.