Evaluation of Multi-Room Sound Insulation Based on Binaural Reproduction and Virtual Reality

As demand increases for automatically analyzing and synthesizing audio information in room acoustics, noise issues from sound transmission in buildings have been gaining more attention. Acoustic intervention, however, is often expensive and requires an in-depth evaluation of its on-site effects. To tackle this issue with acoustic measurements, we considered binaural hearing to demonstrate how people benefit from using both ears instead of one when detecting signals in noise, especially when they are modulated by an interaural phase or level difference. With more audio channels, the less interaural masking level difference, it is challenging to measure the binaural difference in multichannel audio using conventional modeling. In other words, the 3D audio effect increases as more than two channels are involved, but adding channels also impairs the acoustical intelligibility against a background of masking. On the other hand, despite the apparent dissimilarity between visual and aural natures, many techniques used for the visualization of sounds and the auralization of images have been applied. The conventional approach such as the image-source method is limited to a specific listening position or is lacking an intermediate state after calculation. Thus, we developed a system based on the acoustics of encapsulated tracing rays with well-defined position, frequency, phase, intensity, bandwidth, absorption, duration, and behavior in this study. To verify the proposed approach, an audio-visual space is rendered using virtual reality (VR) to evaluate the impacts of acoustic design on buildings. Moreover, a neural network is used to synthesize personalized head-related transfer functions (HRTF) and improve the immersion of binaural auralization in VR. The developed method offered auditory scene decoration. Objective and subjective assessment results confirmed that the accuracy of synthesized HRTFs is enhanced with the linkage between deep modeling and listening perception.