Modelling multi-episodic quality perception for different telecommunication services: First insights

Abstract

Most studies on the perceptual quality evaluation of multimedia services are limited to short media samples or episodes, although typical usage of services occurs in a more regular way over longer periods of exposure (weeks, months or years). So far, it is unclear how the experience of individual episodic use is integrated to form an experience of one service as a whole, spanning over multiple usage episodes. In this paper, we present two studies on the quality experienced during multi-episodic usage of different media and communication services, and first approaches to model multi-episodic quality judgments. An analysis of the data shows that the particular instance of episodic judgment strongly affects the integration function, which makes it difficult to derive models of general validity. Ways to overcome the observed limitations are discussed.

Multi-episodic perceived quality for one session of consecutive usage episodes with a speech telephony service

“…For these experiments, the case of multiple consecutive usage episodes in one session (i.e., up to 45 min overall duration) was selected. This complements initial work on multi-episodic perceived quality in which episodic usage over several days was investigated [10,11,29]. In the two experiments, three aspects were examined: (i) the impact of an increasing number of severely degraded usage episodes, (ii) the impact of varying the temporal position of severely degraded usage episodes, and (iii) the impact of the two different usage situations (i.e., interactive vs. passive use).…”

“…Based upon these findings, Guse et al [11] focused on more severe, unrealistic degradations and increased the number of LP usage episodes per multi-episodic condition. In this experiment, an audio-on-demand service and a video-on-demand service needed to be used once per day over a usage period of 14 days.…”

“…While this model works well for earlier multi-episodic judgments, the prediction accuracy decreases for later multi-episodic judgments. Guse et al [11] could improve the prediction accuracy by accounting for a recency effect using the results of their reported experiment and the results of Möller et al [29]. They evaluated a windowed unweighted average and a linear weighted average.…”

“…Accounting for a peak effect as described by Weiss et al [43] did not improve the prediction accuracy. It must be noted that Guse et al [11] and Möller et al [29] evaluated the prediction accuracy per individual participant. This provides a larger data set compared to deriving a MOS.…”

“…Möller et al [29] and Guse et al [11] evaluated predictors based upon judgments of individual participants. In difference to their approach, the evaluation of potential predictors is conducted here based upon the MOS alone (cf.…”