Agglomerative information bottleneck for speaker diarization of meetings data

Abstract

Abstract. In this paper, we investigate the use of agglomerative Information Bottleneck (aIB) clustering for the speaker diarization task of meetings data. In contrary to the state-of-the-art diarization systems that models individual speakers with Gaussian Mixture Models, the proposed algorithm is completely non parametric . Both clustering and model selection issues of nonparametric models are addressed in this work. The proposed algorithm is evaluated on meeting data on the RT06 evaluation data set. The system is able to achieve Diarization Error Rates comparable to state-of-the-art systems at a much lower computational complexity.2 IDIAP-RR 07-31

Mutual information based channel selection for speaker diarization of meetings data

“…We summarize here the speaker diarization algorithm described in detail in [3]. The clustering steps are described below.…”

“…Previously we have proposed a system [3] based on the Information Bottleneck (IB) principle [4] which is inspired from RateDistortion theory. The speaker diarization aims at finding the clustering that minimize the loss in mutual information between the initial uniform segmentation and the final clustering.…”

Combination of agglomerative and sequential clustering for speaker diarization

“…• Method 1 (aIB): based on conventional aIB+model selection [15]. This method starts with the trivial partition of each element of X in a cluster and performs agglomerative IB clustering until all elements in the space are grouped into a single cluster.…”

“…However, this does not give any further information on the optimal number of clusters which must be estimated with a model selection criterion. In [15] we considered two different model selection metrics: the Minimum Description Length (MDL) and a thresholded Normalized Mutual Information (NMI). NMI can be written as…”

“…In our previous work [15], we introduced a non-parametric diarization system based on the agglomerative information bottleneck (aIB) framework [11]. aIB is a clustering algorithm based on information theoretic framework.…”