An Information Theoretic Approach to Speaker Diarization of Meeting Data

Abstract

Abstract-A speaker diarization system based on an information theoretic framework is described. The problem is formulated according to the Information Bottleneck (IB) principle. Unlike other approaches where the distance between speaker segments is arbitrarily introduced, the IB method seeks the partition that maximizes the mutual information between observations and variables relevant for the problem while minimizing the distortion between observations. This solves the problem of choosing the distance between speech segments, which becomes the Jensen-Shannon divergence as it arises from the IB objective function optimization. We discuss issues related to speaker diarization using this information theoretic framework such as the criteria for inferring the number of speakers, the tradeoff between quality and compression achieved by the diarization system, and the algorithms for optimizing the objective function. Furthermore, we benchmark the proposed system against a state-of-the-art system on the NIST RT06 (Rich Transcription) data set for speaker diarization of meetings. The IB-based system achieves a diarization error rate of 23.2% compared to 23.6% for the baseline system. This approach being mainly based on nonparametric clustering, it runs significantly faster than the baseline HMM/GMM based system, resulting in faster-than-real-time diarization.

Improved overlap speech diarization of meeting recordings using long-term conversational features

“…Speaker diarization system used in the current work is based on a non-parametric bottom-up agglomerative framework [19]. The diarization output assigns each speech segment to a unique cluster (speaker) in the output.…”

“…Overlap labelling reduces the missed speech error [7,8,11]. Table 1 (first row) shows DER of 30.4 for the speaker diarization system without any overlap handling as described in [19]. Let us now compare the results obtained by the baseline overlap detector and the proposed system which exploits information from speaker changes on three tasks overlap exclusion, labelling and both.…”

Speaker diarization of meetings based on large TDOA feature vectors

“…This section briefly summarizes the Information Bottleneck speaker diarization system that operates in a normalized space of relevance variables proposed in [9]. The Information Bottleneck is a distributional clustering technique introduced in [10].…”

“…In order to apply this method to speaker diarization, the set of relevance variables Y = {yn} is defined as the components of a background GMM (M) trained on the entire audio recording [9]. The input to the clustering algorithm is uniformly segmented speech segments xt.…”

Speaker Diarization Based on Intensity Channel Contribution

“…Clustering algorithms are typically based on the Bayesian information criterion (BIC) distance [12] although recent studies have also presented great improvements using other alternatives based on the t-test distance [13]. Most systems extract spectral features related to the spectral envelope such as the Mel frequency cepstral coefficients (MFCCs) [9], [14], although some studies have presented improvements with the fusion of spectral envelope and pitch features [15]. In [16], an exhaustive analysis of the goodness of prosodic and long-term features in speaker diarization is presented.…”