Deep Learning in Speaker Recognition

Abstract

It is supposed in Speaker Recognition (SR) that everyone has a unique voice which could be used as an identity rather than or in addition to other identities such as fingerprint, face, or iris. Even though steps have been taken long ago to apply neural networks in SR, recent advances in computing hardware, new deep learning (DL) architectures and training methods, and access to a large amount of training data have inspired the research community to make use of DL as in a large variety of other signal processing applications. In this chapter, the traditional principle techniques in SR are first briefly reviewed and the potential signal processing aspects of these techniques which can be improved by DL are addressed. Then the recent most successful DL architectures used in SR are introduced and some illustrative experiments from the authors are included.