Using an ordinal ranking rule to find the top-performing Gaussian mixture models for language recognition

In previous work [1], we developed a method for finding the top-performing Gaussian mixture models for the language recognition. This method orders the models from best-performing to worst-performing using calculated dispersion measures. Multiple dispersion measurements are used to produce multiple rankings of the models, which are combined to produce a ranking from which the top-performing models can be extracted. This method has reduced model testing time, since researchers can determine the top-performing models without evaluating the entire population of models. In this paper we demonstrate the ability of our ranking rule to find the top-performing models for different data sets and performance measures. The performance of our ranking rule is also compared to existing ordinal ranking rules: Kohler [2], Arrow & Raynaud [2], Borda [3], and Copeland [3].