INR-Assisted Speculative Spatial Audio Transmission for Zero-Delay Cloud Gaming

Reducing response delay while maintaining high-quality visual and audio output is a critical challenge in cloud gaming services. Speculative transmission has been proposed to reduce the response delay for video frames, whereas the response delay reduction for spatial sound has remained underexplored. In this study, we propose a model-based speculative transmission scheme to deliver high-quality spatial sound with negligible response delay for cloud gaming services. By presharing compressed impulse responses (IRs) between the cloud server and the user’s low-end device using an implicit neural representation (INR) model, our scheme achieves significant traffic reduction while preserving nearly lossless spatial sound quality. In addition, the INR-based model uses a small neural network architecture to reduce the storage cost required for sharing compressed IRs on low-end user devices. Evaluation results using pyroomacoustics show that our scheme achieves a traffic reduction of more than 89.2% while maintaining nearly lossless spatial sound reconstruction.