| dc.description.abstract | Auditory distance perception is fundamental to human life, necessary for spatial orientation,
localization and avoidance of obstacles in environment. However, the psychoacoustics behind
distance perception are quite complex, and particularly problematic for Augmented and Virtual
Reality (AR and VR) applications, which are increasingly becoming an integral part of life.
Acoustical cues such as intensity, direct to reverberant ratio, spectral and binaural differences, as well as source and environment familiarity, are critical to auditory distance perception. However, the manner and extent to which most of these cues contribute is often relative and cumulative, rather than individual and absolute, depending on contextual aspects such as whether the source is in peri-personal, extra-personal or room boundary proximity, but also on environmental factors such as room dimensions, reflectivity and non-idealization. The direct to reverberant ratio may provide an absolute measure of perceptual distance compared to other more relative cues. Since late reverberation remains relatively constant in a room, the early reflections play a critical role for auditory distance. Generally, early reflections and auditory distance perception cues, have been extensively researched in large concert halls. However, listener experience of auditory distance is binaural and takes place in a large variety of small, non-idealised environments, which for the purposes of augmented reality, should be examined from a similar point of view. State of the art solutions rely on complete, physically exact modelling of individual reflections and rooms, to simulate auditory distance. Such approaches are very complex, requiring extensive a priori knowledge about the acoustic environment, and significant computational power. This is not feasible for augmented reality applications, given that the diversity of non-idealized acoustical environments is infinite, computational resources are limited, and no information about the characteristics of the room is provided a priori. In the case of AR, listener familiarity with typical, non-idealized environments facilitates an ongoing, unconscious comparison between physical experience and virtual auditory presentation, which often breaks down as a result of perceived differences. Hence, commercial approaches to auditory distance modelling for AR, rely on relatively coarse and simplistic modification of intensity, reverberation, and filtering. This thesis investigates the direct, plausible simulation of perceptual distance cues using binaural room
impulse response synthesis and generic early reflection patterns, based on a simplified perceptual model for three different, non-idealised, small to medium sized rooms, with minimal a priori details regarding room characteristics. A characterization algorithm was developed to extract objective measures related to auditory distance cues, from a dataset of binaural room impulse responses captured in three, non-idealised rooms of different, small-medium sizes. The resulting characterization informed the development of a Binaural Room Impulse Response (BRIR)
synthesis algorithm, and a matching synthesized BRIR dataset. Objective comparisons between
captured and synthesized BRIRs, found cues such as the Direct to Reverberant Ratio (DRR) to be
decreasing similarly and consistently with distance, except when the source was approaching room boundaries or irregular surfaces such as furniture. At these positions, the natural DRR exhibited inconsistencies, while the simulated DRR maintained consistent decrease with distance. Subjective listening tests using speech and drums stimuli with captured and synthesized BRIRs, suggest that the consistent changes in synthesized DRR also produced a reasonably consistent perception of distance, especially when listeners had no exposure to the room, although this was somewhat also dependent on the stimulus. Nonetheless, given the acoustical variety of non-idealized physical rooms, the challenges in convincingly simulating auditory distance via generic perceptual models remain quite complex and difficult, particularly when computational resources are limited, but even more so when listener familiarity with the room is high, and information about the room is not known a priori, all of which are true for AR. | en |
