Evolutionary minimization of the Rand index for speaker clustering

Abstract

We propose an effective method for clustering unknown speech utterances based on their associated speakers. The method jointly optimizes the generated clusters and the required number of clusters by estimating and minimizing the Rand index. The metric reflects the clustering errors that arise when utterances from the same speaker are placed in different clusters; or when utterances from different speakers are placed in the same cluster. One useful characteristic of the Rand index is that its value only reaches the minimum when the number of clusters is equal to the size of the true speaker population. We approximate the Rand index by a function of the similarity measures between utterances and then use a genetic algorithm to determine the cluster in which each utterance should be located, such that the function is minimized. Our experiment results show that this novel speaker-clustering method outperforms conventional methods that use the Bayesian information criterion to determine the required number of clusters.

Unsupervised Fingerprint Recognition

“…As we found it difficult to represent the purity as a function of the indices, only Rand index is considered as a criterion to find the optimal indices. Adapted from our previous work reported in [18], this study derives an approximate Rand index in consideration of the similarities between fingerprints. Inspect Eq.…”

“…In addition, to minimize the estimated Rand index, we develop two optimization schemes, respectively, based on a genetic algorithm (GA) and tabu search (TS). Compared to [18], in which only GA-based optimization is considered, the clustering methods proposed in this study are more general and flexible. Our experiments conducted using the FVC2002 [13] database but with a subset of the database instead show that the proposed fingerprint clustering method outperforms an intuitive method based on hierarchical agglomerative clustering.…”

“…Here, the index functions by a property that its minimum value occurs only when each cluster consists exclusively of fingerprints from Copyright c 2013 The Institute of Electronics, Information and Communication Engineers only one finger and the number of clusters equals the number of fingers involved in the fingerprint collection. Although the basic strategy of our clustering method stems from [18], the detail of estimating the Rand index in this work are different from that in [18], in order to better accommodate the clustering method to the characteristics of fingerprint. In addition, to minimize the estimated Rand index, we develop two optimization schemes, respectively, based on a genetic algorithm (GA) and tabu search (TS).…”

“…As far as the methodology is concerned, the most closely related research is the biometric clustering in the field of speaker recognition, e.g. [14] and our previous work in [18], which aims to cluster speech utterances based on speaker voice characteristics. This study is inspired by our previous work in [18], which formulates the task of fingerprint clustering as a problem of minimizing the clustering errors characterized by the Rand index.…”

“…[14] and our previous work in [18], which aims to cluster speech utterances based on speaker voice characteristics. This study is inspired by our previous work in [18], which formulates the task of fingerprint clustering as a problem of minimizing the clustering errors characterized by the Rand index. Here, the index functions by a property that its minimum value occurs only when each cluster consists exclusively of fingerprints from Copyright c 2013 The Institute of Electronics, Information and Communication Engineers only one finger and the number of clusters equals the number of fingers involved in the fingerprint collection.…”