Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The Combination of Several Decorrelation Methods to Improve Acoustic Feedback Cancellation

Klaus Linhard, Philipp Bulling

TL;DR

The study tackles bias and slow convergence in a long-path acoustic feedback cancellation system based on a frequency-domain Kalman filter within a multi-delay framework. It introduces and evaluates four decorrelation extensions—variable time delay (vibrato), prediction, non-linear distortion, and a reverberation model—along with a predictor-based speed-up mechanism, all analyzed under practical parameter ranges. Using PESQ and system-distance metrics on public datasets, it shows that each extension improves performance and that combining vibrato with prediction yields the strongest gains, with distortion and reverberation offering additional but smaller benefits. Validation on Lombard speech (Soloducha 2016) and ANIR automotive impulse responses demonstrates the method's potential for robust, real-world acoustic feedback cancellation in challenging environments.

Abstract

This paper extends an acoustic feedback cancellation system by incorporating multiple decorrelation methods. The baseline system is based on a frequency-domain Kalman filter implemented in a multi-delay structure. The proposed extensions include a variable time delay line, prediction, distortion compensation, and a simplified reverberation model. Each extension is analyzed, and a practical parameter range is defined. While existing literature often focuses on a single extension, such as prediction, to describe an optimal system, this work demonstrates that each individual extension contributes to performance improvements. Furthermore, the combination of all proposed extensions results in a superior system. The evaluation is conducted using publicly available datasets, with performance assessed through system distance metrics and the objective speech quality measure PSEQ.

The Combination of Several Decorrelation Methods to Improve Acoustic Feedback Cancellation

TL;DR

The study tackles bias and slow convergence in a long-path acoustic feedback cancellation system based on a frequency-domain Kalman filter within a multi-delay framework. It introduces and evaluates four decorrelation extensions—variable time delay (vibrato), prediction, non-linear distortion, and a reverberation model—along with a predictor-based speed-up mechanism, all analyzed under practical parameter ranges. Using PESQ and system-distance metrics on public datasets, it shows that each extension improves performance and that combining vibrato with prediction yields the strongest gains, with distortion and reverberation offering additional but smaller benefits. Validation on Lombard speech (Soloducha 2016) and ANIR automotive impulse responses demonstrates the method's potential for robust, real-world acoustic feedback cancellation in challenging environments.

Abstract

This paper extends an acoustic feedback cancellation system by incorporating multiple decorrelation methods. The baseline system is based on a frequency-domain Kalman filter implemented in a multi-delay structure. The proposed extensions include a variable time delay line, prediction, distortion compensation, and a simplified reverberation model. Each extension is analyzed, and a practical parameter range is defined. While existing literature often focuses on a single extension, such as prediction, to describe an optimal system, this work demonstrates that each individual extension contributes to performance improvements. Furthermore, the combination of all proposed extensions results in a superior system. The evaluation is conducted using publicly available datasets, with performance assessed through system distance metrics and the objective speech quality measure PSEQ.
Paper Structure (13 sections, 14 equations, 13 figures, 1 table)

This paper contains 13 sections, 14 equations, 13 figures, 1 table.

Table of Contents

  1. Introduction
  2. Structure of an Acoustic Feedback System
  3. The Problem of Reducing Bias
  4. The Problem of Low Convergence Speed
  5. Performance Evaluations
  6. Methods of Bias Reduction
  7. Bias Reduction with Fixed Time Delay
  8. Bias Reduction with Variable Time Delay (Vibrato)
  9. Bias Reduction with Non-linear Distortion
  10. Bias Reduction with a Reverberation Model
  11. Increasing the Convergence Speed
  12. Combination of Several Decorrelation Methods and Data-base Results
  13. Conclusion

Figures (13)

  • Figure 1: Structure of the closed loop feedback system showing the integration of a Kalman filter into a multi-delay frequency domain filter.
  • Figure 2: MOS values for a variable time delay line (vibrato effect) depending on modulation frequency and max. delay time.
  • Figure 3: Convergence of the system distance for $g=0$, 12 and 30 dB. Max. delay is $\pm 2$ msec and modulation frequency is either off, 1, or 2 Hz.
  • Figure 4: Examples for non-linear functions. Top: Signed square; Bottom: Smoothed half-wave rectification with different but fixed values of $c$.
  • Figure 5: THD comparison of curves 1 to 4, for input magnitude range of $x$ from -20 dB to 0 dB (0 dB represents magnitude 1; $\alpha=1$, $\mathit{sc}=1$, cf. Eq. \ref{['eq:scaledmix']}).
  • ...and 8 more figures