Multi-Level Privacy-Preserving Dementia Detection from Speech via Targeted Adversarial Obfuscation and Representation Learning
SoundCryptography and Security
Summary
The authors address privacy concerns in using speech recordings to detect dementia, since these recordings can reveal who is speaking. They created a two-part system that hides the speaker's identity while keeping dementia-related information intact. First, they add noise specifically where keywords appear to confuse speech transcription but keep important speech patterns. Second, they remove speaker-specific features from the data while preserving signs of dementia. Testing showed their method makes it very hard to recognize the speaker, yet still accurately detects dementia.
Authors
Henriette Flore Kenne, Raphael Anaadumba, Mohammad Arif Ul Alam
Abstract
Speech recordings used for dementia detection inherently expose speaker identity, raising critical privacy concerns. Existing methods typically address only singular threats and fail to resolve the privacy--utility trade-off. We propose a multi-level framework designed to neutralize two distinct eavesdropping vectors. At the signal level, a Cumulative Signal Attack (CSA) concentrates perturbations in keyword-aligned regions to maximize transcription error (Word Error Rate WER = 1.00) while preserving vital prosodic biomarkers. At the feature level, a Gradient Reversal Layer (GRL) with Mutual Information (MI)-guided noise injection suppresses speaker-discriminative dimensions while retaining dementia-relevant diagnostic structure. Evaluated on the DementiaBank Pitt Corpus, our framework achieves near-chance speaker identification (Equal Error Rate EER = 0.59, F1 = 0.003) while maintaining strong dementia classification performance (F1 = 0.78, AUC = 0.86).