Speaker
Description
Not all reflections in measured Ambisonic room impulse responses (ARIRs) arrive in distinct time intervals. In principle, compact spherical microphone arrays for higher-order Ambisonic recordings enable the decomposition of coincident reflections.However, existing approaches for directionally decomposing and upmixing higher-order ARIRs resolve fewer coincident reflections than the (𝑁+1)² components of an order-𝑁 ARIR would theoretically permit.They either rely on a fixed spatial sectorization (e.g., HO-SIRR), which may be insufficiently resolved or misaligned, or estimate a limited number of time-varying directions (e.g., REPAIR), whose associated signal extraction can become ill-conditioned when directions are not well separated.This contribution adopts the Ambisonic Multi-Direction Decomposition Method (AMDDM) to address these limitations in higher-order ARIR decomposition. It employs a fixed maximum-determinant grid that yields (𝑁+1)² well-separated directional signals, whose short-term covariance matrix is used to align the grid with the acoustic reflections by penalizing off-diagonal entries.In a numerical study with simulated ARIRs containing multiple reflections coinciding within a single time frame, we compare AMDDM with HO-SIRR and REPAIR. The results highlight the conditions under which AMDDM improves directional decomposition, as well as its limitations.