SpeechGuard - Recoverable and Customizable Speech Privacy Protection

Explore a 20-minute conference presentation introducing SpeechGuard, an innovative system for protecting speech privacy in cloud environments while maintaining recoverability and customization options. Learn how this research addresses the critical challenge of uploading speech data to cloud servers without compromising both acoustic and content privacy. Discover the system's dual approach using a multi-parameter warping function with inverse capabilities for voice conversion to protect acoustic information, combined with automatic and manual detection mechanisms for encrypting sensitive text content. Understand how SpeechGuard categorizes listeners into permission groups with varying levels of access through assigned warping parameters and encryption keys, enabling personalized recovery of speech data based on user permissions. Examine experimental results from three datasets demonstrating superior performance compared to baseline systems in anonymity, sensitive content confidentiality, and attack resistance, while providing the flexibility to define tailored privacy requirements and protection strength levels.