Binaural rendering technology over loudspeakers and headphones

Dingding Yao, Huaxing Xu, Junfeng Li, Risheng Xia, Yonghong YanNippon 2020 · DOI

Abstract

To provide listeners with an immersive listening experience, binaural rendering technology has become an important research topic, especially with the rising prominence of virtual/augmented reality. This paper introduces recent advancements in binaural rendering over headphones and loudspeakers. We present an improved free-field crosstalk cancellation (CTC) method using a stochastic robust approximation to handle listener movement. Additionally, we propose a parametric elevation control approach that models key perceptual cues—spectral peaks and notches—using digital filters. Subjective and objective experiments verify the effectiveness of these algorithms in maintaining spatial stability and audio quality.

1. Introduction: The Challenges of Binaural Audio

Binaural rendering aims to replicate the 3D auditory experience of the real world. However, two primary technical hurdles persist:

  1. The "Sweet Spot" Problem in Loudspeakers: Crosstalk occurs when the left ear hears the right speaker's signal and vice versa. While Crosstalk Cancellation (CTC) can mitigate this, traditional methods are fragile; even minor head movements can cause the spatial illusion to collapse.
  2. The "Flat World" Problem in Elevation: Simulating height (elevation) is difficult because it relies on highly individualized spectral cues from the outer ear (pinna). Using generic models often results in "spectral coloration," making audio sound unnatural or metallic.

2. Methodology

The researchers developed three distinct approaches to address these challenges.

2.1 Improved Free-Field Method (IFM)

Instead of relying on complex, individualized Head-Related Transfer Functions (HRTFs), the authors modeled the human head as a simple rigid sphere. This "Improved Free-Field Method" formulates the head's effect using an attenuation factor and a phase difference factor.

Figure 1: Geometry of the Improved Free-Field Method (IFM) for crosstalk cancellation, modeling the listener's head as a rigid sphere.

2.2 Stochastic Robust Approximation

To improve robustness against head movement, the listener's position is treated as a random variable within a Stochastic Least Mean Square (LMS) framework. This creates a "softer" filter that maintains channel separation even when the listener is not perfectly centered.

2.3 Parametric Elevation Control

By analyzing the CIPIC HRTF database, the team found that spectral "notches" (frequency dips) move predictably with elevation, while "peaks" remain stable. They used 2nd-order digital filters to dynamically shift these notches based on learned regression rules, effectively "adding" height to horizontal audio signals.

3. Experimental Validation

The suggested algorithms were tested using:

  • Subjective Panels: 11 participants for localization and MUSHRA (Multiple Stimulus with Hidden Reference and Anchors) tests, and 9 participants for elevation experiments.
  • Stimuli: A diverse range of sounds including pink noise, speech, music, gunshots, and guitar.

4. Results and Discussion

The proposed methods demonstrated significant improvements in both objective metrics and subjective perception.

4.1 Localization and Robustness

In simulations of a 10° speaker setup, the stochastic method improved channel separation by 21.99% over traditional methods when the listener's head was misaligned by 4cm. Subjective localization tests showed that the IFM maintained high accuracy across various azimuthal angles.

Figure 2: Subjective localization results showing perceived angle versus azimuthal angle for different rendering methods (FM, IFM, HFM).

4.2 Audio Quality and Elevation

In MUSHRA tests, the IFM scored significantly higher than generic HRTF methods, indicating lower spectral coloration. For elevation, listeners accurately perceived intended heights, though performance slightly decreased at extreme angles (above 67.5°).

5. Conclusion

This research demonstrates that simplified physical models (like the rigid sphere) and stochastic optimization can outperform complex generic models in practical binaural rendering. Future work will focus on reducing front-back confusion and integrating these methods into real-time VR/AR systems.

Summary updated

Abstract

To provide listeners with an immersive listening experience, binaural rendering technology has become an important research topic, especially with the rising prominence of virtual/augmented reality in recent years. In this paper, we introduce our recent works on binaural rendering technology over headphones and loudspeakers. The first work is on crosstalk cancellation (CTC), which is critical for loudspeaker-based binaural rendering. An improved free-field CTC method is first presented, in which the head effects are formulated using an attenuation factor and a phase difference factor. A stochastic robust approximation method is then suggested to further improve its robustness against the perturbation perception caused by the listener's head movement or rotation. The second work is on elevation perception of sound images, which is critical for headphone- and loudspeaker-based binaural rendering methods. By analyzing the elevation-dependent head-related transfer function (HRTF), a parametric elevation control approach is presented in which the key perceptual cues (i.e., spectral peaks and notches) are modeled using digital filters and controlled according to learned rules. The effectiveness and performance of the suggested algorithms are verified by subjective and objective experiments.
BibTeX
@inproceedings{yao2020binaural,
  author = {Dingding Yao and Huaxing Xu and Junfeng Li and Risheng Xia and Yonghong Yan},
  title = {{Binaural rendering technology over loudspeakers and headphones}},
  booktitle = {Nippon},
  year = {2020},
  doi = {10.1250/ast.41.134},
}
Figure