From Speech to Signals: Audio Dementia Research and Applications
Overview
This topic covers research translating speech features into measurable signals that help detect, monitor, and understand dementia. It spans acoustic analysis, machine learning models, clinical validation, and practical applications (screening tools, remote monitoring, and research platforms).
Key components
- Acoustic features: prosody (pitch, intonation), speech rate, pauses, articulation, spectral features, voice quality (jitter, shimmer).
- Linguistic features: vocabulary richness, syntactic complexity, semantic coherence, word-finding pauses.
- Signal-processing methods: noise reduction, feature extraction (MFCCs, formants), temporal segmentation.
- Machine learning: supervised classifiers, deep learning (CNNs, RNNs, transformers), feature selection, explainability methods.
- Clinical validation: cross-sectional vs longitudinal studies, ground truth via neuropsychological tests and biomarkers, sensitivity/specificity metrics.
- Ethical & practical considerations: consent, data privacy, demographic biases, deployment in diverse populations.
Typical study pipeline
- Participant recruitment and consent.
- Audio collection (structured tasks, spontaneous speech, phone/telehealth