Implementing Higher-Order Ambisonics (HOA) on consumer devices is hindered by their sparse, irregular microphone arrays, which challenge conventional methods with issues like spatial aliasing and ill-conditioning. This paper proposes a unified Spherical Harmonic Beamforming (SHB-AE) framework that recasts HOA encoding as a spatial filtering problem, enabling robust, signal-independent solutions. We develop two approaches: a frequency-domain (FD) method with compensation for high-frequency artifacts, and a time-domain (TD) methodthat holistically optimizes broadband FIR filters for enhanced stability. The framework is inherently scalable, allowing on-demand order expansion. Using a measured smartphone array, comprehensive objective and subjective evaluations demonstrate the clear superiority of theTD method. It excels in signal fidelity, spatial accuracy, and temporal consistency, outperforming baseline and FD approaches. The TD method also maintains its advantage in adverse conditions, showing remarkable robustness against noise, reverberation, and multi-source environments.It provides a practical, high-performance pathway for enabling high-fidelity spatial audio capture on ubiquitous consumer devices without requiring complex signal analysis or large datasets.

Individualized head-related transfer functions (HRTFs) require accurate pinna geometry, yet commodity multi-view captures leave the ear region self-occluded and weakly textured. We present a practical pipeline that couples ear-centric acquisition with 3D Gaussian splatting (3DGS) and the boundary element method (BEM) for complete HRTF computation. The protocol augments horizontal views with per-ear elevated captures under directional lighting; 3DGS training with depth-distortion regularization yields watertight meshes via truncated signed distance function (TSDF) fusion. Standardized head coordinates and ear-canal annotations interface the mesh with BEM. Experimental evaluations demonstrate that our method achieves lower ear-region geometric error and lowerfull-band spectral distortion compared to existing image-based personalized reconstruction baselines including AudioEar, NeuS, and Metashape MVS.

Binaural rendering is typically assessed via timbre and localization accuracy, while its intrinsic spatial resolution remains rarely quantified. This paper proposes a perceptual evaluation method based on Minimum Audible Angle (MAA) measurements to estimate the azimuthal just-noticeable difference (JND) introduced by binaural rendering algorithms. We systematically compared several rendering algorithms across eight reference azimuths using two participant-allocation paradigms. The results show that spatial resolution is significantly influencedby Ambisonic order and choice of the rendering algorithm, with MAA thresholds systematically decreasing as the truncation order increases. Furthermore, the proposed method successfully captures physiological spatial characteristics and identifies resolution limits imposed by reference angles. While both participant-allocation paradigms yield consistent qualitative trends, the repeated-measures design provides superior data stability. These findings demonstrate that the proposed MAA-based method is an effective tool for quantifying the spatial resolutionof binaural rendering algorithms.

While Neural Audio Codecs (NAC) have revolutionized monaural audio compression, achieving high-fidelity dual-channel coding at low bitrates remains a significant challenge. Existing approaches often rely on naive independent channel quantization, leading to phase incoherence, or entangled latent modeling, which sacrifices spatial precision for spectral energy. This paper proposes a novel dual-channel coding framework based on contentspatial disentanglement. Reframing spatial reconstruction as an informed source separation task, our architecturesynergizes a frozen, pre-trained DAC encoder for robust mono content preservation with a parameter-efficient side information encoder that predicts fine-grained time-frequency masks. To ensure precise spatial imaging, we introduce explicit physical constraints into the end-to-end training. Experimental results indicate that at low bitrates of 9 and 11 kbps, the proposed method outperforms state-of-the-art dual-mono neural baselines and industry standards in both objective spatial metrics and subjective MUSHRA evaluations.

Sound source localization and identity tracking are fundamental tasks in acoustic scene analysis, enabling machines to determine what, where, and when sound events occur. While deep attractor-based networks have demonstrated improved performance under an unknown number of sources, maintaining continuous source tracking over longform audio remains challenging due to memory limitations and permutation ambiguities across adjacent segments. In this paper, we propose a Recursive Attractor Network (RANet) for long-form sound source localizationand identity tracking with a variable number of sources. RANet explicitly represents attractors as transferable embeddings and recursively propagates them across adjacent audio segments using a LSTM-based model, thereby preserving source identity continuity over time. Experimental results on simulated datasets demonstrate that RANet achieves robust long-form localization and consistent source identity tracking, outperforming baseline approaches. 

Higher-Order Ambisonics (HOA) encoding from sparse, irregular microphone arrays remains a critical challenge for consumer spatial audio capture in immersive communication and XR. We propose Flow-HOA, a generative framework that jointly optimizes a multi-dimensional objective encompassing time-domain, spectral, and spatial fidelity while producing a deployable, time-invariant bank of Finite Impulse Response (FIR) encoding filters. Using conditional flow matching, the model learns to map a simple prior distribution to the target distribution of FIR filtercoefficients. Training is guided by a composite loss that balances time-domain waveform fidelity, multi-resolution spectral consistency, sub-band energy preservation, and spatial directivity constraints. Objective evaluations on synthetically simulated data demonstrate improved performance over strong model-based baselines in both signal fidelity and spatial accuracy metrics. Subjective listening tests on real microphone array recordings further confirmthat Flow-HOA yields higher overall sound quality with reduced artifacts, demonstrating generalization from synthetic training data to real-world capture conditions.

Achieving sound field reproduction (SFR) with high sound quality and accurate spatial localization in automotive cabins is particularly challenging due to complex acoustics and constrained loudspeaker layouts. This paper proposes a learning-based method to address this challenge, integrating a spatial domain physics-informed constraint based on plane-wave decomposition (PWD) with a multi-position control strategy. Results from both objective evaluations and in-situ subjective listening tests consistently validated the superiority of the proposed approach over several baseline methods. Moreover, we show that the correlation of spatial power maps (SPMs) derived from PWD provides a reliable objective metric that closely reflects perceived spatial localization in the cabin environment.

Sound source localization (SSL) under unknown or variable sound sources conditions remains a challenging task. Existing methods suffer from limitations such as grid resolution constraints, fixed output dimensionality and insufficient exploitation of mutual assitance between temporal and spatial information. In this paper, we propose an Envelope Separation Aided Multi-Task Learning model for blind source counting and localization, which adaptively generates attractors to estimate source numbers and jointly optimizes envelope separation and direction estimation through a multi-task learning model using permutation invariant training (PIT). Experimental results demonstrated that the proposed model achieved better performance, by leveraging temporal domain envelope separation to aid spatial localization, outperforming baseline approaches.