Text this: Dissimilarity-based classification for stochastic models of embedding spaces applied to voice pathology detection