Subjective assessment of codec transparency has traditionally sought to minimize the roles of listeners and audio programs in the computation of codec transparency measures by limiting their variability where possible, e.g. by insisting on -expert- listeners. By separately modeling listener and program effects, the Rasch Model is not constrained in this way, making it possible to examine relationships between measured listener -severity- in rating codecs and such variables as musical and audio background, gender, age, and the effect of repeated listening. The authors also explore relationships between the measured -intolerance- of audio programs to codec artifacts and such variables as reverberance, dynamic range, crest factor, and high frequency density. These exploratory analyses are based on tests of five codecs for Lucent Technologies.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.