Privacy-preserving speaker verification using secure binary embeddings

Remote speaker verification services typically rely on the system having access to the users recordings, or features derived from them, and/or a model for the users voice. This conventional approach raises several privacy concerns. In this work, we address this privacy problem in the context of a speaker verification system using a factor analysis based front-end extractor, the so-called i-vectors. Preserving privacy in our context means that neither the system observes voice samples or speech models from the user, nor the user observes the universal model owned by the system. This is achieved by transforming speaker i-vectors to bit strings in a way that allows for the computation of approximate distances, instead of exact ones. The key to the transformation uses a hashing scheme known as secure binary embeddings. Then, an SVM classifier with a modified kernel operates on the hashes. Experiments showed that the secure system yielded similar results as its non-private counterpart. The approach may be extended to other types of biometric authentication.