A Hybrid Beamforming-Post-Filter Framework for Ambisonics Enhancement

Ambisonics recordings encoded from multi-microphone signals enable effective capture of three-dimensional soundfield information, thus have been widely adopted in immersive media, virtual reality, and augmented reality applications. However, in real-world scenarios, the directly acquired Ambisonics signals typically contain not only the desired source component but also competing interferers, reverberation, and ambient noise, posing significant challenges for downstream spatial audio processing tasks. This paper targets an Ambisonics-based multi-input multi-output (MIMO) framework to extract the components of interest while preserving their inherent spatial cues. To address this challenge, we propose a three-stage enhancement framework, each responsible for: (i) initial Ambisonics enhancement by extending the traditional maximum directivity beamformer to a MIMO formulation in the spherical harmonics domain; (ii) refined enhancement by applying a post-filter, with the required power spectral density estimated using an oblique projection method; (iii) achieving a balance between signal quality and spatial cue preservation by a mixing strategy that combines the beamformed output and the post-filtered signal. Extensive experiments conducted under challenging acoustic conditions involving reverberation, background noise, and interfering sources validate the superiority of the proposed framework over benchmark methods.