Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Decomposing the Influence of Physical Acoustic Modeling on Neural Personal Sound Zone Rendering: An Ablation Study

Hao Jiang, Edgar Choueiri

TL;DR

A controlled ablation study on a head-pose-conditioned binaural PSZ renderer using the Binaural Spatial Audio Neural Network to guide prioritization of measurements and models when constructing training ATFs under limited budgets.

Abstract

Deep learning-based Personal Sound Zones (PSZs) rely on simulated acoustic transfer functions (ATFs) for training, yet idealized point-source models exhibit large sim-to-real gaps. While physically informed components improve generalization, individual contributions remain unclear. This paper presents a controlled ablation study on a head-pose-conditioned binaural PSZ renderer using the Binaural Spatial Audio Neural Network (BSANN). We progressively enrich simulated ATFs with three components: (i) anechoically measured frequency responses of the particular loudspeakers(FR), (ii) analytic circular-piston directivity (DIR), and (iii) rigid-sphere head-related transfer functions (RS-HRTF). Four configurations are evaluated via in-situ measurements with two dummy heads. Performance metrics include inter-zone isolation (IZI), inter-program interference (IPI), and crosstalk cancellation (XTC) over 100-20000 Hz. Results show FR provides spectral calibration, yielding modest XTC improvements and reduced inter-listener IPI imbalance. DIR delivers the most consistent sound-zone separation gains (10.05 dB average IZI/IPI). RS-HRTF dominates binaural separation, boosting XTC by +2.38/+2.89 dB (average 4.51 to 7.91 dB), primarily above 2 kHz, while introducing mild listener-dependent IZI/IPI shifts. These findings guide prioritization of measurements and models when constructing training ATFs under limited budgets.

Decomposing the Influence of Physical Acoustic Modeling on Neural Personal Sound Zone Rendering: An Ablation Study

TL;DR

A controlled ablation study on a head-pose-conditioned binaural PSZ renderer using the Binaural Spatial Audio Neural Network to guide prioritization of measurements and models when constructing training ATFs under limited budgets.

Abstract

Deep learning-based Personal Sound Zones (PSZs) rely on simulated acoustic transfer functions (ATFs) for training, yet idealized point-source models exhibit large sim-to-real gaps. While physically informed components improve generalization, individual contributions remain unclear. This paper presents a controlled ablation study on a head-pose-conditioned binaural PSZ renderer using the Binaural Spatial Audio Neural Network (BSANN). We progressively enrich simulated ATFs with three components: (i) anechoically measured frequency responses of the particular loudspeakers(FR), (ii) analytic circular-piston directivity (DIR), and (iii) rigid-sphere head-related transfer functions (RS-HRTF). Four configurations are evaluated via in-situ measurements with two dummy heads. Performance metrics include inter-zone isolation (IZI), inter-program interference (IPI), and crosstalk cancellation (XTC) over 100-20000 Hz. Results show FR provides spectral calibration, yielding modest XTC improvements and reduced inter-listener IPI imbalance. DIR delivers the most consistent sound-zone separation gains (10.05 dB average IZI/IPI). RS-HRTF dominates binaural separation, boosting XTC by +2.38/+2.89 dB (average 4.51 to 7.91 dB), primarily above 2 kHz, while introducing mild listener-dependent IZI/IPI shifts. These findings guide prioritization of measurements and models when constructing training ATFs under limited budgets.
Paper Structure (14 sections, 14 equations, 3 figures, 1 table)

This paper contains 14 sections, 14 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Method
  3. Neural PSZ Rendering Framework
  4. Physically Informed ATF Modeling
  5. Cumulative Ablation Protocol
  6. Metrics
  7. Experimental Setup and Evaluation
  8. Physical Setup
  9. Training Conditions
  10. Measurement-Based Evaluation Protocol
  11. Results and Analysis
  12. Overall measured performance
  13. Incremental contribution of FR, DIR, and RS-HRTF
  14. Conclusion

Figures (3)

  • Figure 1: Geometric configuration of the experimental testbed. The playback system features a 24-element loudspeaker array mounted on a rigid baffle, comprising two rows: 8 woofers ($100$--$2000$ Hz) and 16 tweeters ($2$--$20$ kHz). Two HATS are positioned symmetrically at a distance of 1.0 m from the array with lateral offsets of $\pm 0.5$ m relative to the centerline, oriented perpendicular to the array midpoint.
  • Figure 2: Broadband performance summary for the four cumulative configurations. Top row: log-mean metric values over 100--20,000 Hz for Listener 1 and Listener 2 under C0--C3. Bottom row: incremental changes under the cumulative ablation protocol (FR: C1$-$C0; DIR: C2$-$C1; RS-HRTF: C3$-$C2).
  • Figure 3: Frequency-dependent IZI, IPI, and XTC for the four cumulative training configurations, evaluated at a single static pose. Legends correspond to C0 (baseline point-source simulation), C1 (+FR), C2 (+FR+DIR), and C3 (+FR+DIR+RS-HRTF). Top row: Listener 1. Bottom row: Listener 2.