Listener movement prediction based realistic real-time binaural rendering

Abstract

Realistic binaural rendering refers to a process that implies applying the physically correct signals to the eardrums of a listener. In real-time binaural rendering systems, used in VR environments, the accurate signals should be appropriately updated as the listener freely moves in a simulated room. The update process requires room impulse response and head-related transfer function for each new position and pose of the listener. The classical approach uses a filter set, estimated for a particular position. The switching (to a new position-related filter set) is available after the current signal analysis frame is fully processed. In this paper, we propose an approach for immediate switching to the correct filtered binaural output with a rotation resolution of 3 degrees. Our solution is based on pre-filtering applied to the upcoming audio signal frame. The proposed solution does not exceed the latency limit required by VR audio systems and is suitable for quick head movements during the rendering process.