Organizations: Center of Intelligent Acoustics and Immersive Communications, School of Artificial Intelligence, Northwestern Polytechnical University, Xi’an, 710072, China · Digital Signal Processing Lab, School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore
Ambisonics delivers compact scene based spatial audio representation, yet higher order Ambisonic encoding poses difficulties for wearables and embedded hardware. Their microphone arrays are often sparse, irregular, and constrained by device specific boundary conditions. These factors make the spherical-harmonic (SH) domain encoding ill conditioned: inverse filtering amplifies noise, while deterministic neural encoders may overfit to array-specific responses or smooth ambiguous higher-order components. This paper presents DiffM2A, a geometry-adaptive conditional diffusion framework for robust Ambisonic encoding from sparse MAs with variable topologies. Its Geometry-Adaptive Spherical Harmonic Projection (GASHP) front-end constructs boundary-aware SH steering functions and applies an energy-normalized modal projection, mapping array-dependent observations to a common modal representation without explicit pseudo-inverse computation. A dual-branch Elucidated Diffusion Model then estimates complex Ambisonic coefficients, conditioned on both the raw microphone spectra and GASHP features. Sound intensity and rotational equivariance losses further enhance inter-channel phase consistency and structured behavior across SH subspaces. Evaluations on both first- and second-order Ambisonic encoding tasks, using simulated room-acoustics and real-world LOCATA recordings, demonstrate that DiffM2A outperforms conventional and neural baseline methods on signal fidelity, spectral accuracy, spatial coherence, and binaural cue preservation. Additional experiments show that these gains are largely retained across unseen five-microphone layouts and under mismatched open-array and rigid-sphere boundary models.
Figures & tables
Figure 1: Overview of the proposed DiffM2A framework. During training, the target Ambisonic coefficients aN are corrupted with Gaussian noise to obtain aNσ . During inference, the coefficient estimate is initialized from Gaussian noise and iteratively denoised using DPM-Solver.
Task
Method
Single Source
Dual Source
SI-SDR
Coh
Mag. Err
ILD.Err
SI-SDR
Coh
Mag. Err
ILD.Err
(dB) ↑
↑
(dB) ↓
(dB) ↓
(dB) ↑
↑
(dB) ↓
(dB) ↓
FOA
Param.
7.82
0.695
5.93
3.29
4.24
0.630
8.85
4.50
AmbiSpatial
10.57
0.629
8.99
4.91
6.83
0.551
12.32
5.78
Gen-A
11.09
0.649
8.81
4.74
7.59
0.566
11.57
5.59
Attention-based
12.43
0.654
7.90
3.82
8.76
0.593
10.23
4.97
Table 1: Objective evaluation results on the simulated dataset for single- and dual-source scenarios.
Figure 2: Comparison of mean magnitude spectrum error and coherence metrics under the simulated dual-source SOA configuration.
Task
Method
Single Source
Dual Source
SI-SDR
Coh
Mag. Err
ILD.Err
SI-SDR
Coh
Mag. Err
ILD.Err
(dB) ↑
↑
(dB) ↓
(dB) ↓
(dB) ↑
↑
(dB) ↓
(dB) ↓
FOA
Param.
0.85
0.71
5.82
3.88
-0.81
0.60
7.15
4.62
AmbiSpatial
3.82
0.58
4.34
3.31
1.98
0.47
5.58
4.12
Gen-A
4.65
0.63
3.92
2.95
2.85
0.52
5.12
3.61
Attention-based
5.21
0.68
3.65
2.68
3.42
0.58
4.68
3.24
Table 2: Objective evaluation results on the LOCATA dataset for single- and dual-source scenarios.
Method
SI-SDR (dB) ↑
Mag.Err (dB) ↓
Ground Truth
6.86
3.71
Param.
1.01
7.84
AmbiSpatial
3.94
5.43
Gen-A
4.71
4.97
Attention-based
5.48
4.53
Proposed
8.05
3.14
Table 3: Comparison of sensor-domain reprojection errors under the dual-source SOA configuration.
Configuration
SI-SDR (dB) ↑
Coh ↑
Mag. Err (dB) ↓
ILD.Err (dB) ↓
LEDM
6.75
0.224
11.52
6.08
LEDM+LIV
7.32
0.288
10.64
5.62
LEDM+Lrot
8.54
0.332
10.08
5.38
LEDM+LIV+Lrot
9.74
0.405
9.27
5.02
w/o GASHP
7.16
0.272
10.92
5.75
GASHP + U-Net
7.21
0.282
10.97
5.87
Table 4: Ablation study for dual-source SOA encoding.
Figure 3: Comparison of STFT spectrograms of the ground-truth HOA coefficients (top) and the corresponding GASHP-projected modal representations (bottom) for four representative harmonic channels ( W , Y , b2,−2 , and b2,0 ) under the simulated dual-source configuration.
Method
C-5-6
C-5-4
L-5-4
L-5-3
Avg.
Params (M)
Param.
2.42
3.26
1.35
1.51
2.14
–
AmbiSpatial
4.53
5.13
4.40
4.51
4.64
8.12
Gen-A
5.48
6.05
5.21
5.34
5.52
9.10
Attention-based
7.54
7.73
7.15
7.23
7.41
1.19
w/o GASHP
7.66
7.75
7.16
7.31
7.47
8.53
Proposed
9.81
10.10
9.54
9.65
9.78
8.66
Table 5: Comparison on unseen arrays in SI-SDR (dB).
Figure 4: Time-domain waveforms of the reconstructed W -channel under the simulated dual-source SOA configuration.
Figure 5: Comparison of STFT spectrograms for reconstructed Ambisonic channels ( W , Y , b2,−2 , and b2,0 ) across different methods under the simulated dual-source configuration.
Method
Training Cost
Inference Time
Parametric
–
0.35 s
AmbiSpatial
Medium
1.23 s
Gen-A
Medium
1.36 s
Attention-based
Low
0.93 s
Proposed
High
2.66 s
Table 6: Comparison of the training cost and average inference time (evaluated per 4-second audio sample at 16 kHz). The neural network-based models are evaluated on a single NVIDIA GeForce RTX 5090 GPU, whereas the parametric baseline is executed on an AMD EPYC 9554 64-Core CPU.
We address the problem of encoding room impulse responses (RIRs) into high-order Ambisonics (HOA) representations from arbitrary and potentially insufficient or incomplete microphone array measurements. This task is fundamentally ill-posed for microphone arrays with limited spatial capture capabilities, such as irregular or sparse arrays, as classical linear methods fail to reconstruct high-order spatial detail. We introduce a diffusion-based generative framework that models the statistical properties of HOA RIRs. This enables device-agnostic encoding from arbitrary microphone arrays, potentially unseen during data measurement. Our approach incorporates a posterior sampling procedure that enforces consistency between the estimated signals and the measurements while plausibly reconstructing spatial information that is unobservable from the limited measurements alone. Experiments on simulated data demonstrate that our method outperforms linear and neural baselines, achieving accurate HOA RIR estimation up to 12th order. A listening test with binaural renderings, including both simulated and measured RIRs, further confirms that the proposed method yields higher perceptual similarity to reference Ambisonics RIRs than all baselines. The flexibility and accuracy of the proposed framework opens new possibilities for scalable acoustics simulations.
Eloi Moliner, Christoph Hold, Juan Azcarreta Ortiz +5
Reality Labs Research, Meta · Acoustics Lab, DICE, Aalto University, Espoo, Finland
With the rapid development of virtual reality (VR) and augmented reality (AR), spatial audio recording and reproduction have gained increasing research interest. Higher Order Ambisonics (HOA) stands out for its adaptability to various playback devices and its ability to integrate head orientation. However, current HOA recordings often rely on bulky spherical microphone arrays (SMA), and portable devices like smartphones are limited by array configuration and number of microphones. We propose SHB-AE, a spherical harmonic beamforming based method for Ambisonics encoding using a smartphone microphone array (SPMA). By designing beamformers for each order of spherical harmonic functions based on the array manifold, the method enables Ambisonics encoding and up-scaling. Validation on a real SPMA and its simulated free-field counterpart in noisy and reverberant conditions showed that the method successfully encodes and up-scales Ambisonics up to the fourth order with just four irregularly arranged microphones.
Yuhuan You, Yufan Qian, Tianshu Qu +2
Peking University · Beijing Xiaomi Mobile Software Co., Ltd · Xiaomi Communications Co., Ltd
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 filter coefficients. 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 confirm that Flow-HOA yields higher overall sound quality with reduced artifacts, demonstrating generalization from synthetic training data to real-world capture conditions.
Yuhuan You, Yufan Qian, Tianshu Qu +2
State Key Laboratory of General Artificial Intelligence, School of Intelligence Science and Technology, Peking University, Beijing, China · 2Beijing Xiaomi Mobile Software Co., Ltd · 3Xiaomi Communications Co., Ltd