Can latent speech quality dimensions be quantified directly?

Previous studies revealed that the quality of transmitted speech can well be described by three perceptual dimensions: “discontinuity”, “noisiness”, and “coloration”. In this paper, a method is presented for quantifying these dimensions directly in subjective tests. Two experiments were conducted according to this method with large sets of test conditions. A detailed description of the test procedure and a thorough analysis of the results is presented. The proposed method is reliable and produces meaningful and orthogonal results for diverse stimuli sets, in particular for stimuli containing multi-dimensional degradations. The method might form the basis for collecting the data for future diagnostic speech quality estimators.