一种基于物理特性和感知特性的混响模拟方法

Risheng Xia, Junfeng Li, HuaXing Xu, Yonghong YanScientia 2015 · DOI

Abstract

Reverberation simulation in many areas including live music and computer games has important applications. Traditional physical and perceptual-based reverberation simulation techniques require a high computational cost and lack the specific characteristics of the room, respectively. This paper presents a hybrid artificial reverberation synthesis method (HPPR): the early reflections are generated by convolving with the impulse response modeled by the image source method (ISM) based on the physical characteristics; the late reverberation is produced by the feedback delay network (FDN) based on the perceptual characteristics. Furthermore, we pro-pose a method to estimate the energy decay relief (EDR) of the modeled early reflection and thus compute the parameters of the feedback delay network using the method. This process results in the smooth transition from the early reflections to the late reverberation in the time-frequency domain. Numerical simulation and subjective hearing experiments confirm the effectiveness of the proposed method.

Introduction

Creating realistic artificial reverberation is a fundamental challenge in audio engineering, requiring a balance between physical accuracy and computational efficiency. Physical methods, such as the Image Source Method (ISM), provide high fidelity by modeling specific room geometries but suffer from exponential computational growth during late reverberation. Conversely, perceptual methods like Feedback Delay Networks (FDN) are computationally efficient for real-time applications but often lack the unique acoustic signature of a specific physical space.

A significant hurdle in hybrid systems is the "seam"—the transition point where early reflections end and late reverberation begins. Inconsistent transitions in volume or spectral balance can lead to audible artifacts. This paper introduces the HPPR (Hybrid Physically- and Perceptually-based Reverberation) framework to address these issues through a novel parametric model.

Methods

The HPPR framework integrates physical modeling with perceptual simulation through three primary components:

  1. Early Reflection Modeling: A low-order ISM is utilized to generate the initial reflections, incorporating air absorption models to ensure natural high-frequency decay over distance.
  2. Late Reverberation Modeling: An FDN is employed for efficient late-stage simulation. To ensure the FDN matches the specific room characteristics, its parameters (such as $T_{60}$ decay times) are dynamically tuned.
  3. Energy Decay Relief (EDR) Estimation: This is the core innovation. The authors derived a closed-form mathematical expression to predict the EDR based on room dimensions and wall reflection coefficients. This allows for the instant calculation of FDN parameters, ensuring a seamless "hand-off" between the two stages.
Block diagram of the HPPR framework showing the integration of ISM, EDR estimation, and FDN.

Results and Discussion

The effectiveness of HPPR was validated through numerical simulations and subjective listening tests.

Perceptual Quality and Efficiency

In MUSHRA-style subjective tests involving 10 participants and various audio sources (speech, song, piano), HPPR scored nearly identically to the full ISM "gold standard." This indicates that listeners could not distinguish the hybrid approach from the more computationally intensive physical model. Furthermore, HPPR achieved a approximately 30% computational speedup compared to previous hybrid methods like ISM-DRM.

Seamless Transition

A critical result of the EDR estimation method is the smoothness of the transition at the crossover time ($t_c$). Time-frequency analysis confirms the absence of energy spikes or spectral "holes," demonstrating that the parametric model successfully aligns the late reverb tail with the early reflections' spectral characteristics.

Time-frequency magnitude plot showing the smooth transition between early reflections and late reverberation at time $t_c$.

Conclusion

The HPPR method successfully combines the strengths of physical and perceptual reverberation modeling. By utilizing a novel EDR estimation technique, it achieves a seamless transition in the time-frequency domain while maintaining real-time efficiency. Future work will focus on extending the model to non-rectangular room geometries and exploring dynamic acoustic environments where sources and listeners are in motion.

Summary updated

Abstract

Reverberation simulation in many areas including live music and computer games has important applications. Traditional physical and perceptual-based reverberation simulation techniques require a high computational cost and lack the specific characteristics of the room, respectively. This paper presents a hybrid artificial reverberation synthesis method (HPPR): the early reflections are generated by convolving with the impulse response modeled by the image source method (ISM) based on the physical characteristics; the late reverberation is produced by the feedback delay network (FDN) based on the perceptual characteristics. Furthermore, we pro-pose a method to estimate the energy decay relief (EDR) of the modeled early reflection and thus compute the parameters of the feedback delay network using the method. This process results in the smooth transition from the early reflections to the late reverberation in the time-frequency domain. Numerical simulation and subjective hearing experiments confirm the effectiveness of the proposed method.
BibTeX
@inproceedings{xia2015,
  author = {Risheng Xia and Junfeng Li and HuaXing XU and Yonghong Yan},
  title = {{一种基于物理特性和感知特性的混响模拟方法}},
  booktitle = {Scientia},
  year = {2015},
  doi = {10.1360/n112014-00326},
}
Figure