Title: Robust Speaker Clustering in Eigenspace

Authors: Robert Faltlhauser, Guenther Ruske

Abstract: In this paper we propose a speaker clustering scheme working in 'Eigenspace'. Speaker models are transformed to a low-dimensional subspace using 'Eigenvoices'. In this subspace simple speaker distance measures (Euklid) can be applied. Clustering can be accomplished with base models (for Eigenvoices) like GMMs as well as conventional HMMs. In case of HMM models re-projection to original space readily yields acoustic models. Clustering in subspace produces well-balanced cluster and is easily to control. In the field of speaker adaptation several principal techniques can be distinguished. The most prominent among them are Bayesian adaptation (MAP), transformation based approaches (MLLR) as well as so-called Eigenspace (Eigenvoices) techniques. The latter have become popular, as they make use of a-priori information about the distribution of speaker models. Speaker clustering is a further attractive adaptation scheme, especially since it can easily be combined with the other methods.

a01rf086.ps a01rf086.pdf