Privacy-preserving speaker verification using garbled GMMS

In this paper we present a privacy-preserving speaker verification system using a UBM-GMM technique. Remote speaker verification services rely on the system having access to the user's recordings, or features derived from them, and a model representing the user's voice. 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. Our approach uses Garbled Circuits for obtaining an implementation that simultaneously is secure, has high accuracy and is efficient. To the best of our knowledge this is the first privacy-preserving speaker verification system that accomplishes all these three goals.