As immersive technologies such as Virtual Reality (VR) and Augmented Reality (AR) become increasingly prevalent, the demand for efficient, perceptually optimized audio compression has grown rapidly. Traditional audio codecs prioritize signal fidelity over perceptual relevance and often fail to preserve critical spatial cues under low-bitrate constraints. In this paper, we propose a novel perceptual neural audio codec tailored for low-bandwidth immersive applications, with an emphasis on preserving both auditory salience and spatial integrity. Our approach combines psychoacoustic feature extraction with a deep encoderdecoder architecture, conditioned on spatial metadata and trained using perceptual and spatially aware loss functions. The codec is evaluated across multiple datasets including speech, music, and ambient scenes, demonstrating superior performance over existing classical and neural codecs in terms of both objective metrics (e.g., SNR, LSD, ViSQOLAudio) and subjective Mean Opinion Scores (MOS). Moreover, the lightweight model architecture enables real-time decoding on mobile hardware, making it suitable for deployment in VR/AR headsets, mobile devices, and low-power audio systems. This work establishes a foundation for scalable, high-quality spatial audio delivery under severe bandwidth constraints, advancing the state of the art in perceptual audio coding for immersive media.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=23088
(272KB)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
