Modeling of HR Filters for Audio Object Rendering

The Immersive Voice and Audio Services (IVAS) codec supports encoding of audio objects, represented as mono audio streams coupled with metadata describing the position (and orientation) of the objects. The format is called Independent Streams with Metadata (ISM). The audio objects may be rendered to a binaural output format, to be played back by headphones, by means of convolving the decoded audio streams of each object with a head-related filter for the listener-relative Direction of Arrival (DOA) of the audio object. These filters are generated using a novel model-based approach, which allows each object to be rendered at any position around the listener. The model is continuous over the DOA domain, based on continuous B-spline basis functions, and the model parameters can be trained on any input head-related filter set. The model reduces the storage compared to the full head-related filter set and improves the rendering quality by smoothing out potential measurement errors in the filter data.