acoustic scene simulation

Current speech separation and enhancement models exhibit limited generalization under mobile-source scenarios, primarily due to insufficient diversity and realism in evaluation data—both real-world and synthetic datasets fail to adequately reflect practical acoustic conditions. To address this, we propose SonicSim: the first customizable acoustic simulation framework specifically designed for mobile sound sources. SonicSim leverages Habitat-sim for physically accurate, multi-source spatial modeling and integrates LibriSpeech, FSD50K, and FMA audio corpora with Matterport3D 3D indoor environments. Based on this framework, we construct SonicSet—a large-scale, high-fidelity benchmark dataset—and complement it with real-world counterpart recordings. Experiments demonstrate that models trained on SonicSet achieve significantly improved generalization on real mobile-source recordings compared to those trained on existing synthetic datasets, effectively narrowing the synthetic-to-real acoustic domain gap.

Audio algorithm evaluation is often hindered by limited scale, high cost, and insufficient diversity of real-world measurements. To address this, this work proposes replacing physical measurements with room acoustics simulation and systematically compares wave-based numerical simulation against geometric acoustics simulation for evaluating audio signal processing (ASP) and acoustic modeling (AML) algorithms. Leveraging multiple simulation engines—including one wave-based solver and two geometric simulators—alongside ground-truth measurements from real rooms, we conduct comparative evaluations across three representative audio algorithms. Results demonstrate that wave-based simulation achieves significantly higher fidelity than geometric methods (with substantially lower average error relative to measurements), efficiently generating diverse, high-fidelity acoustic scenes. This markedly improves test coverage and reproducibility. To our knowledge, this is the first study to empirically validate wave-based simulation as a low-cost, high-fidelity standard evaluation tool, establishing a new paradigm for robust, scalable audio algorithm development and benchmarking.

This work introduces a novel task—material-controllable room impulse response (RIR) generation—aiming to synthesize high-fidelity acoustic responses dynamically, conditioned on user-specified material configurations (e.g., floor, wall finishes) and multimodal audio-visual observations of indoor scenes. To support this, we present Acoustic Wonderland, the first acoustic dataset enabling fine-grained material combinations and synchronized multi-view audio-visual recordings. We further propose a new audio-visual–material fusion encoder-decoder architecture that explicitly models material properties and their geometric-acoustic mapping. Experiments demonstrate substantial improvements over existing baselines and state-of-the-art methods in RIR prediction accuracy, material sensitivity, and generation diversity. Notably, our approach enables real-time, interactive editing of material parameters during inference—a capability unprecedented in prior acoustic simulation frameworks.

To address the low efficiency and poor reproducibility in generating binaural datasets for spatial audio research, this paper introduces Binamix—an open-source Python library built upon the SADIE II database (HRIRs/BRIRs from 20 subjects). Binamix supports multi-channel layouts, parametric spatial positioning, and large-scale synthesis. Its core innovation is an improved HRIR/BRIR interpolation method integrating enhanced Delaunay triangulation, enabling high-fidelity binaural rendering at arbitrary azimuth and elevation angles. The framework is modular, fully reproducible, and designed for downstream tasks—including codec evaluation and model training—featuring integrated example scripts, visualization tools, and utility functions. Released under the Apache 2.0 license, Binamix significantly advances standardization and development efficiency in spatial audio research.

Existing replay attack detection methods rely on single-channel recordings and exhibit poor generalization across unseen acoustic environments. Method: We propose a multi-channel spatial cue–based detection enhancement framework. First, we construct an acoustic simulation system that incorporates empirically measured loudspeaker directivity patterns—novel in replay attack research—alongside room impulse responses, multi-channel convolution, and noise injection to generate high-fidelity multi-channel synthetic data for both reverberant and anechoic spoofing scenarios. Second, we design the M-ALRAD detector, which explicitly exploits spatial features from microphone arrays. Contribution/Results: This work establishes the first physically interpretable, multi-channel simulation paradigm specifically tailored for replay attack detection. Experiments demonstrate that our framework significantly improves model robustness and generalization performance under unseen environmental conditions, outperforming prior single-channel approaches.

This work proposes a deep learning–based covariance matrix upsampling method to enhance the spatial resolution of acoustic imaging using a four-element tetrahedral microphone array under hardware-constrained conditions. The key innovation lies in the novel integration of frequency-dynamic convolution with a 2D convolutional neural network, which effectively captures the spatial–frequency coupling structure and frequency-dependent characteristics inherent in the covariance matrix. Experimental results demonstrate that the proposed approach reduces the root mean square error (RMSE) of sound source localization to 0.432, significantly outperforming a random baseline (RMSE = 0.548). Moreover, the reconstructed beamforming heatmaps closely approximate those produced by a ground-truth 32-channel spherical array, highlighting the method’s efficacy in high-fidelity acoustic imaging with minimal hardware resources.

This study addresses the inflated accuracy often reported by data-driven models in predicting room acoustic parameters, which stems largely from biases in evaluation protocols—particularly the overestimation of performance when test locations lack actual measurements. To rectify this, the work proposes a consistent evaluation framework that explicitly distinguishes between “location interpolation” and “prediction at truly unknown locations.” Using multi-condition measured data, it systematically evaluates three approaches: random forests, hybrid CNNs, and inverse distance weighting. Results show that high predictive performance (R² = 0.80–0.88) is achievable only when measured impulse responses at test locations are available as positional fingerprints. Under realistic generalization conditions—without any test-point data—performance drops substantially (R² = 0.09–0.57), though learning-based models still demonstrate practical advantages in predicting sound strength and reverberation time. This work underscores the dominant influence of evaluation protocols on reported metrics and establishes a more reliable benchmark for acoustic modeling.

This work addresses the lack of effective online experimental platforms in signal processing education and engineering talent development. To bridge this gap, the authors developed and have continuously refined J-DSP, a web-based simulation environment that pioneered the migration of the original Java-based DSP toolkit to an HTML5 architecture, enabling cross-platform— including mobile—accessibility. The platform integrates advanced topics such as digital filter design, FFT-based spectral analysis, machine learning for signal classification, and quantum Fourier transform. Having operated reliably for 25 years, J-DSP has been widely adopted in university courses and National Science Foundation–funded programs, including REU, IRES, and RET initiatives, significantly advancing the modernization of signal processing pedagogy and fostering STEM workforce development.

This work addresses the challenge that existing audio generation methods, relying on a single viewpoint, struggle to maintain spatial consistency of sound in 3D scenes as the user moves. To overcome this limitation, the authors propose Scene2Sound, a training-free framework that achieves viewpoint-invariant spatially consistent soundscapes by identifying sound-emitting objects through auditory anchoring and binding them to persistent 3D locations. The approach integrates vision-language models, 3D Gaussian Splatting rendering, and multi-view geometric matching, introducing Gaussian set matching for coherent multi-view sound source fusion. Two novel metrics are designed to evaluate spatial consistency. Experiments demonstrate that Scene2Sound significantly outperforms single-view baselines in both synthetic and real-world 360° reconstructed 3DGS scenes while preserving high audio quality, with user studies confirming its perceptual advantages.

This work addresses the lack of a unified, high-quality transcoding method for spatial audio across diverse acquisition formats—such as Ambisonics or microphone arrays—and arbitrary playback systems. The authors propose a general parametric framework that estimates spatial metadata of primary sources and ambient sound in the time–frequency domain, constructs a spatial covariance model tailored to the target playback setup, and derives an optimal linear downmix matrix. This approach supports independent rotation between acquisition and playback geometries and, for the first time, unifies processing for both Ambisonics and raw microphone array inputs. It accommodates arbitrary array configurations, variable numbers of sources, and arbitrary angular power distributions of ambient sound. Listening tests demonstrate that the method significantly outperforms existing parametric renderers across various content types and playback configurations, with particularly notable perceptual improvements for low-order or geometrically constrained arrays.