Perceptual detection thresholds for alterations of the azimuth of early room reflections

Sammanfattning

Spatial audio, that is, the reproduction of an acoustic event including all spatialinformation, has become a major research topic in recent years. The computergames industry, for example, is looking for ways to reproduce the created spaces asrealistically as possible, not only visually but also acoustically. A particularly largearea of interest in spatial audio is the authentic reproduction of a room. The questionarises how far the reproduced angles of a reflection can deviate from the actual anglewithout the room being perceived differently. This has already been investigatedfor single reflections, but how and when several shifted reflections together cause aperception change is still largely unknown.Therefore, in this work it is investigated how far the azimuth angles of the firstfive reflections of a room impulse response have to be shifted, so that the originalroom impulse response can just be distinguished from the manipulated one. For thispurpose, 20 subjects performed a headphone-based adaptive ABX threshold test inwhich the selected reflections, initially all coming from one side, were shifted closerto or further away from their original positions at each adaptation step until theindividual point of just noticeable difference was found. For this, the number ofsteps for moving each reflection from the adaptation start to the original positionwas the same, but the step size differed due to the individual angle differences.The recordings of a drum and a male speaker were used in the test. The resultsshow that the five selected reflections for both test signals can be shifted between4.33 and 17.13 , depending on the step size of the reflection, until a difference isaudible. The perceived differences between the rooms were described as changesof the center of gravity in the room or the localization of the sound source. Nomajor difference between the median values of the drum and the speech signal couldbe found. The inexperienced listening test participants were mainly responsible forall outliers found in the threshold tests while the median thresholds were slightlyhigher, suggesting that the reproduced angles of the reflections could potentiallydeviate more for the broad society. An additional ABX test comparing the originalroom with a room where the first reflections were mirrored showed a clear perceptualdifference.