Impact of Compression on the Performance of the Room Impulse Response Interpolation Approach to Spatial Audio Synthesis

In Virtual Reality (VR) an application where a user is able to move freely in the virtual or physical environment, the 3D sound scene reproduced over headphones, including the acoustic Room Impulse Response (RIR) must be constantly updated. Storing and transmitting the RIR for every possible source and listening position is impractical and hence efficient compression is required. This paper investigates compression and interpolation approaches for synthesising the RIR at new listening positions using a discrete set of compressed directional RIRs (DRIRs) recorded with an Ambisonics B-format microphone. The predicted sound quality and localisation accuracy of Ambisonic sound scenes derived using interpolated, compressed RIRs is based on the Neural Similarity Index Measure (NSIM) applied to the time-frequency phase information derived from reference and synthesised B-format channels and as used in the AMBIQUAL objective spatial audio quality metric. Results show a strong correlation between the compression bitrate and the objective quality scores of B-format Ambisonics audios including interpolated and ground truth DRIRs.