Paper 2019/976

Towards real-time hidden speaker recognition by means of fully homomorphic encryption

Martin Zuber, Sergiu Carpov, and Renaud Sirdey


Securing Neural Network (NN) computations through the use of Fully Homomorphic Encryption (FHE) is the subject of a growing interest in both communities. Among different possible approaches to that topic, our work focuses on applying FHE to hide the model of a neural network-based system in the case of a plain input. In this paper, using the TFHE homomorphic encryption scheme, we propose an efficient fully homomorphic method for an argmin computation on an arbitrary number of encrypted inputs and an asymptotically faster - though levelled - equivalent scheme. Using these schemes and a unifying framework for LWE-based homomorphic encryption schemes (Chimera), we implement a very time-wise efficient, homomorphic speaker recognition scheme using the neural-based embedding system VGGVox. This work can be generalized to all other similar Euclidean embedding-based recognition systems. While maintaining the best-of-class classification rate of the VGGVox system, we implement a speaker-recognition system that can classify a speech sample as coming from one of a 100 hidden model speakers in less than one second.

Available format(s)
Publication info
Preprint. Minor revision.
Fully Homomorphic EncryptionNeural NetworksLWENearest Neighbour problemembedding-based recognition systemsspeaker-recognitionFHE performance
Contact author(s)
zubersmartin @ gmail com
sergiu carpov @ cea fr
renaud sirdey @ cea fr
2019-08-29: last of 2 revisions
2019-08-29: received
See all versions
Short URL
Creative Commons Attribution


      author = {Martin Zuber and Sergiu Carpov and Renaud Sirdey},
      title = {Towards real-time hidden speaker recognition by means of fully homomorphic encryption},
      howpublished = {Cryptology ePrint Archive, Paper 2019/976},
      year = {2019},
      note = {\url{}},
      url = {}
Note: In order to protect the privacy of readers, does not use cookies or embedded third party content.