Using voice to generate cryptographic keys

In this position paper, we motivate and summarize our work on repeatably generating cryptographic keys from spoken user input. The goal of this work is to enable a device to generate a key (e.g., for encrypting files) upon its user speaking a chosen password (or passphrase) to it. An attacker who captures the device and extracts all information it contains, however, should be unable to determine this key. We outline our approach for achieving this goal and present preliminary empirical results for it. We also describe several directions for future work.