You are the key: Generating cryptographic keys from voice biometrics

In this work we apply randomized biometric templates (RBTs) to voice biometrics by performing an experiment using speech samples from the TI46 database. Additionally, we present a novel algorithm for extracting reliable features from voice biometrics and analyze the resulting entropy of the cryptographic keys generated by the RBT algorithm. We evaluate our implementation by analyzing the number of guesses required by a powerful adversary to generate a user´s cryptographic key when given access to the user´s decrypted template and population statistics. Furthermore, we compare our results to the results of prior work. We demonstrate that RBTs are able to generate cryptographic keys with at least 30 bits of entropy for 36% of the population and at least 40 bits of entropy for 7% of the population, while keys generated using prior work only contain at least 20 bits of entropy for 19% of the population. We also demonstrate that RBT generated keys are able to achieve a maximum entropy of 51 bits, while keys generated using prior work are only able to achieve a maximum entropy of 26 bits.