Early-time sound field reconstruction from binaural room impulse responses and perceptual evaluation in auditory augmented reality

Abstract Realistic, coherent auditory augmented reality (AAR) requires virtual auditory illusions to be spatialised and acoustically matched to the local environment. To realise AAR, the early part of the sound field with the direct sound and early reflections are reconstructed from a sparse set of binaural room impulse responses (BRIRs). A system-inversion-based BRIR reconstruction method is proposed, based on a sparse spatio-temporal representation of the sound field. Late reflections are taken from an available measurement and treated as position and orientation independent for reproduction. The resulting BRIRs are perceptually evaluated in AAR scenarios and compared against ground truth simulations. The reconstruction was conducted using both the head-related impulse responses (HRIRs) used to generate the data, thus assuming exact knowledge of the HRIR of the listener, and a non-individualised one. The objective evaluation indicates that reconstructed sound fields preserve critical spatial and room acoustic cues, remaining close to the ground truth. Consistent with this, two dynamic binaural listening experiments show that the reconstructed and ground truth sound fields are perceived to be similar. A source was simulated based on the measured directivity pattern of a loudspeaker and evaluated in two orientations: facing the listening area and oriented towards a wall. The experiments were conducted under low-reverberant conditions. The direct-facing source showed no significant perceptual advantage with position and orientation-dynamic early reflections. However, when the source was rotated towards the wall, dynamic early reflections became perceptually important. Using a non-individualised HRIR for system inversion and reproduction had minimal impact on perceptual quality, suggesting the method may remain effective beyond individualised reproduction scenarios.