Generalized variogram is used to extract vocal dysperiocities in disordered speech produced by dysphonic speakers. Both signal and dysperiodicity are passed through a filter bank and a segmental signalto- dysperiodicity ratio is defined in each frequency band. Multivariate analysis is carried out to summarize the degree of perceived hoarseness. The predictor variables are the segmental signal-todysperiodicity ratios in the different bands. It is shown that high correlations are achieved by linear regression analysis of the segmental signal-to-dysperiodicity ratios in different non-overlapping frequency bands.