ClaritySpeech: Dementia Obfuscation in Speech

2025-07-12

AI Summary

Dementia induces atypical speech patterns, exacerbating communication difficulties and increasing risks of speaker identity or clinical condition disclosure; however, conventional ASR systems exhibit poor robustness to such non-canonical speech. This paper proposes a fine-tuning-free, low-data-dependent voice de-identification framework that innovatively integrates ASR transcription, semantic-aware text perturbation, and zero-shot TTS reconstruction—preserving speaker identity while obfuscating dementia-associated linguistic markers. Evaluated on the ADReSS and ADReSSo datasets under adversarial conditions, the method incurs only a 10–16% drop in average F1 score, retains ~50% speaker similarity, substantially reduces WER, and improves MOS speech quality from 1.65 to 2.15. The approach thus achieves a balanced trade-off among privacy preservation, intelligibility, and naturalness.

Technology Category

• Natural Language Processing: Speech
• Machine Learning: Privacy
• Application Domains: Humanities & Computational Social Science

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Dementia, a neurodegenerative disease, alters speech patterns, creating communication barriers and raising privacy concerns. Current speech technologies, such as automatic speech transcription (ASR), struggle with dementia and atypical speech, further challenging accessibility. This paper presents a novel dementia obfuscation in speech framework, ClaritySpeech, integrating ASR, text obfuscation, and zero-shot text-to-speech (TTS) to correct dementia-affected speech while preserving speaker identity in low-data environments without fine-tuning. Results show a 16% and 10% drop in mean F1 score across various adversarial settings and modalities (audio, text, fusion) for ADReSS and ADReSSo, respectively, maintaining 50% speaker similarity. We also find that our system improves WER (from 0.73 to 0.08 for ADReSS and 0.15 for ADReSSo) and speech quality from 1.65 to ~2.15, enhancing privacy and accessibility.

Problem

• Corrects dementia-affected speech patterns for better communication
• Preserves speaker identity in low-data environments
• Improves speech quality and privacy for dementia patients

Innovation

• Integrates ASR, text obfuscation, and zero-shot TTS
• Corrects dementia speech without fine-tuning
• Preserves speaker identity in low-data environments