Identification of perceptive dimensions of speech and audio codecs subjective quality

New generations of speech and audio codecs have several complex types of impairment. To assess their quality, telecommunication laboratories perform subjective and/or objective assessments. Anchor signals are required to get reliable subjective assessment and comparable results of studies performed at different moments or results of studies from different laboratories. The design of these signals needs a description of the impairments useful to find out their correlated physical characteristics. Assuming that codecs quality is multidimensional, the codecs were projected into a perceptive space whose principal axes represented their main impairments. We carried out a verbalization task to label these main dimensions. A Multiple Factor Analysis highlighted a four-dimensional perceptive space. Two of these dimensions were modelled and validated. The proposed reference signals allowed covering the defaults of two different coding techniques.