Provable Consent for Voice User Interfaces

The proliferation of acoustic human-computer interaction raises privacy concerns since it allows Voice User Interfaces (VUI) to overhear human speech and to analyze and share content of overheard conversation in cloud datacenters and with third parties. This process is non-transparent regarding when and which audio is recorded, the reach of the speech recording, the information extracted from a recording and the purpose for which it is used. To return control over the use of audio content to the individual who generated it, we promote intuitive privacy for VUIs, featuring a lightweight consent mechanism as well as means of secure verification (proof of consent) for any recorded piece of audio. In particular, through audio fingerprinting and fuzzy cryptography, we establish a trust zone, whose area is implicitly controlled by voice loudness with respect to environmental noise (Signal-to-Noise Ratio (SNR)). Secure keys are exchanged to verify consent on the use of an audio sequence via digital signatures. We performed experiments with different levels of human voice, corresponding to various trust situations (e.g. whispering and group discussion). A second scenario was investigated in which a VUI outside of the trust zone could not obtain the shared secret key.