Stereo Audio Rendering for Personal Sound Zones Using a Binaural Spatially Adaptive Neural Network (BSANN)

TL;DR

The paper addresses personal sound zone PSZ limitations under head motion by introducing the Binaural Spatially Adaptive Neural Network (BSANN), an ear-wise extension of the SANN that jointly optimizes the acoustic field at both ears of multiple head-tracked listeners via filters $g(omega)$. It employs a two-stage training with physically informed ATFs—combining anechoic loudspeaker responses, analytic piston directivity, and rigid-sphere HRTFs—and an explicit active crosstalk cancellation stage to boost isolation and binaural fidelity. Empirical results show that BSANN yields more balanced inter-listener isolation (IZI/IPI) and substantial passive and active XTC gains compared with SANN, with further improvements from physically informed ATFs and XTC. The work demonstrates a robust framework for stereo and binaural PSZ rendering in realistic environments, with practical implications for multi-listener listening spaces and head-tracked audio systems.

Abstract

A binaural rendering framework for personal sound zones (PSZs) is proposed to enable multiple head-tracked listeners to receive fully independent stereo audio programs. Current PSZ systems typically rely on monophonic rendering and therefore cannot control the left and right ears separately, which limits the quality and accuracy of spatial imaging. The proposed method employs a Binaural Spatially Adaptive Neural Network (BSANN) to generate ear-optimized loudspeaker filters that reconstruct the desired acoustic field at each ear of multiple listeners. The framework integrates anechoically measured loudspeaker frequency responses, analytically modeled transducer directivity, and rigid-sphere head-related transfer functions (HRTFs) to enhance acoustic accuracy and spatial rendering fidelity. An explicit active crosstalk cancellation (XTC) stage further improves three-dimensional spatial perception. Experiments show significant gains in measured objective performance metrics, including inter-zone isolation (IZI), inter-program isolation (IPI), and crosstalk cancellation (XTC), with log-frequency-weighted values of 10.23/10.03 dB (IZI), 11.11/9.16 dB (IPI), and 10.55/11.13 dB (XTC), respectively, over 100-20,000 Hz. The combined use of ear-wise control, accurate acoustic modeling, and integrated active XTC produces a unified rendering method that delivers greater isolation performance, increased robustness to room asymmetry, and more faithful spatial reproduction in real acoustic environments.

Figures (6)

• Figure 1: Overview of the proposed two-stage BSANN framework. Step 1 performs PSZ pretraining to obtain the frequency-domain loudspeaker filters $\mathbf{g}(\omega_n)$ under a loss function composed of the bright-zone and dark-zone objectives together with the gain and compactness constraints. Step 2 refines the filters to $\mathbf{g}_{\mathrm{XTC}}(\omega_n)$ under a loss function dominated by the active XTC term and regularized by teacher-anchoring, small-weight BZ/DZ protector terms, and the gain and compactness constraints.
• Figure 2: Processing pipeline used to construct the physically informed acoustic transfer functions (ATFs) for BSANN training. Simulated room impulse responses (RIRs) are decomposed into direct and reflected components, transformed into frequency-domain room transfer functions (RTFs), and combined with measured anechoic responses, analytic piston directivity, and rigid-sphere HRTFs to obtain the ATFs used in the BSANN loss functions.
• Figure 3: Physical experimental setup. The loudspeaker system consists of a 24-element array mounted at listener height, with 8 woofers in the lower row and 16 tweeters in the upper row. Two head-and-torso simulators (B&K HATS) were positioned symmetrically about the array center, each placed 0.5 m to the left or right and 1.0 m in front of the loudspeaker array, as illustrated in the top-view diagram.
• Figure 4: Comparison between the monophonic SANN-PSZ model (blue) and the BSANN-PSZ model trained with ideal point-source ATFs and without physically informed components or active XTC (red). Each panel shows the log-frequency–weighted IZI, IPI, or XTC curve for Listener 1 and Listener 2.
• Figure 5: Comparison between a BSANN-PSZ model trained with ideal point-source ATFs (blue) and a BSANN-PSZ model trained with physically informed ATFs (red), with both models trained without active XTC. Each panel shows the log-frequency–weighted IZI, IPI, or XTC curve for Listener 1 and Listener 2.