SilentWhisper: inaudible faint whisper speech input for silent speech interaction

Author

Hiraki, Hirotaka and Rekimoto, Jun

Abstract

Voice interaction has become an integral part of our daily digital experience, from controlling smart homes to accessing AI assistants. However, privacy concerns and social considerations severely limit voice interface adoption in public spaces. While silent speech interfaces promise a solution, existing approaches require user-specific training data, support limited vocabularies, or demand intrusive sensors in contact with the user's face. We present SilentWhisper, which enables private voice interaction through ultra-low volume whispered speech that is inaudible beyond 30cm while maintaining high recognition accuracy. Using a headset microphone and deep learning, our system achieves 97.7% word recognition accuracy across a vocabulary of 454 words without requiring per-user training. We demonstrate that SilentWhisper enables unobtrusive voice interaction while preserving privacy. Our approach represents a significant advancement in making voice interfaces practical for sensitive information and public spaces.