Un modello additivo per matrici asimmetriche di confusione

Confusion matrices,widely used in audiometrics, show asymmetries wich become stronger in conditions of filtering or masking of the sound. Such asymmetries were usually neglected, and the matrices were forced to be symmetric a priori. In this paper a model is presented to study the causes of such asymmetries,under the hypothesis that these asymmetries are due to additive distortion terms, correlated to the frequencies of occurrences of the concerning phonemes in the spoken language. A calculus based on statistics of occurences in spoken Italian confirms the utility of the additive model and also of the hypoteses; goodness of fit is specified.