Rank aggregation of dispersion measure orderings for estimating Gaussian mixture model language recognition performance

Gaussian mixture models with shifted delta ceptra features are known to provide high-performance language recognition. The performance of the models is typically assessed using measurements derived from detection estimation tradeoff (DET) curves, which is a costly process. This paper describes a new method for estimating Gaussian mixture model performance which reduces the need for the performance measurements. This new methodology uses dispersion measures combined with rank aggregation to order models from best-performing to worst-performing. This ranking is used to identify the top-performing N% models, which allows researchers to train models and categorize them by performance. This method reduces model testing, since researchers can select categories of models they choose to evaluate.