For Human Ears Only: Preventing Automated Monitoring on Voice Data

As voice communication becomes an essential part of modern life, the exposure of sensitive information through audio calls presents significant privacy risks. Malicious actors can gain access to this data by compromising user devices, exploiting communication channels, or attacking data servers, making it vulnerable to automated monitoring systems that can identify speakers and extract speech content. To address these privacy concerns, we introduce VoiceSecure, the first microphone module designed to prevent automated monitoring of speech while preserving its natural sound for humans. By leveraging the principles of human auditory perception, VoiceSecure employs a set of speech modifications that are adaptively tuned in real-time to obscure speaker identity and speech content, without compromising the quality of the audio for human listeners. This hardware-based solution mitigates the risk of software-based attacks, integrating seamlessly with commercial devices via audio jack or Bluetooth. Comprehensive evaluation across the state-of-the-art speaker verification and speech recognition models, and a variety of speech datasets demonstrates that VoiceSecure outperforms traditional methods of protecting speech from automated monitoring while keeping it intelligible for humans