Multidimensional scaling and preference mapping were used for the perceptual analysis of the quality of speech corrupted by car noise in mobile communications. 41 processing chains, representing, e.g., transmission of speech over mobile networks, were studied. 30 screened subjects were used in the quality test and 15 screened and trained subjects in the MDS test. Based on an external profiling of the auditory characteristics, the dimensions appeared to relate to general naturalness of speech, limitation of the frequency band of speech, smoothness of speech and noisiness of speech. The Phase IV, ideal point model was used to predict the quality with an average error of about 6 %, to study the interaction between the attributes and the linearity of the attributes.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.