The dissertation at hand deals with aspects of quality perception ofinteractive audiovisual application systems of moderate complexity as e.g.defined in the MPEG-4 standard. Because in these systems the availablecomputing power is limited, it is decisive to know which factors influencethe perceived quality. Only then can the available computing power bedistributed in the most effective and efficient way for the simulation anddisplay of audiovisual 3D scenes. Whereas quality factors for the unimodalauditory and visual stimuli are well known and respective models ofperception have been successfully devised based on this knowledge, this isnot true for bimodal audiovisual perception. For the latter, it is onlyknown that some kind of interdependency between auditory and visualperception does exist. The exact mechanisms of human audiovisual perceptionhave not been described. It is assumed that interaction with an applicationor scene has a major influence upon the perceived overall quality.The goal of this work was to devise a system capable of performingsubjective audiovisual assessments in the given context in a largelyautomated way. By applying the system, first evidence regarding audiovisualinterdependency and influence of interaction upon perception should becollected. Therefore this work was composed of three fields of activities:the creation of a test bench based on the available but (regarding theaudio functionality) somewhat restricted MPEG-4 player, the preoccupationwith methods and framework requirements that ensure comparability andreproducibility of audiovisual assessments and results, and the performanceof a series of coordinated experiments including the analysis andinterpretation of the collected data. An object-based modular audiorendering engine was co-designed and -implemented which allows to performsimple room-acoustic simulations based on the MPEG-4 scene descriptionparadigm in real-time. Apart from the MPEG-4 player, the test benchconsists of a haptic Input Device used by test subjects to enter theirquality ratings and a logging tool that allows to journalize all relevantevents during an assessment session. The collected data can be exportedcomfortably for further analysis using appropriate statistic tools.A thorough analysis of the well established test methods andrecommendations for unimodal subjective assessments was performed to findout whether a transfer to the audiovisual bimodal case is easily possible.It became evident that - due to the limited knowledge about the underlyingperceptual processes - a novel categorization of experiments according totheir goals could be helpful to organize the research in the field.Furthermore, a number of influencing factors could be identified thatexercise control over bimodal perception in the given context.By performing the perceptual experiments using the devised system, itsfunctionality and ease of use was verified. Apart from that, some firstindications for the role of interaction in perceived overall quality havebeen collected: interaction in the auditory modality reduces a human'sability of correctly rating the audio quality, whereas visually based(cross-modal) interaction does not necessarily generate this effect.
展开▼
