Speaker identification by anchor models with PCA/LDA post-processing

Speaker representation by location is a new technique of speaker recognition and adaptation. It consists in representing a new speaker not in an absolute manner, but relatively to a set of well trained speaker models. Each new speaker is represented by its location in an optimal representation space. This paper addresses the location task. It describes a representation space built either by clustering speakers or by selecting an optimal subset of them. In this representation space, speaker location is then performed by the anchor models technique to find vector of coordinates. An orthogonalization process is then applied to the vector of coordinates, so as to compute the distance properly. This orthogonalization process (PCA or LDA) proves experimentally to improve significantly the recognition.