An ordinal ranking method for estimating Gaussian mixture model language recognition performance

We developed a new method for estimating the language recognition performance of Gaussian mixture models. This new method calculates dispersion measures for models, then orders the models from best-performing to worst-performing using them. We use multiple dispersion measurements to produce multiple rankings of the models. We produce a compromise ranking among the dispersion measure orderings, and use this ranking to identify the top-performing N% models. This method reduces the number of models needing evaluation, since researchers can select categories of models to test in lieu of evaluating the entire population of models. This paper presents a new ordinal ranking rule that produces a compromise ranking that identifies the top-performing N% models with 100% recall. We also compare the performance of this new ranking rule to existing ordinal ranking rules: Kohler, Arrow & Raynaud, Borda, and Copeland.