Comparative Study of Acoustic Echo Cancellation Algorithms for Speech Recognition System in Noisy Environment

TL;DR

This paper compares various acoustic echo cancellation algorithms, focusing on their performance in noisy environments, convergence, and computational complexity, to identify the most effective approaches for speech recognition systems.

Contribution

It provides a comprehensive comparison of multiple AEC algorithms, highlighting their strengths and weaknesses in different noisy and dynamic conditions.

Findings

Frequency domain algorithms show faster convergence.

Sparse adaptive algorithms improve performance in sparse systems.

Variable step-size algorithms enhance robustness in noisy environments.

Abstract

Traditionally, adaptive filters have been deployed to achieve AEC by estimating the acoustic echo response using algorithms such as the Normalized Least-Mean-Square (NLMS) algorithm. Several approaches have been proposed over recent years to improve the performance of the standard NLMS algorithm in various ways for AEC. These include algorithms based on Time Domain, Frequency Domain, Fourier Transform, Wavelet Transform Adaptive Schemes, Proportionate Schemes, Proportionate Adaptive Filters, Combination Schemes, Block Based Combination, Sub band Adaptive Filtering, Uniform Over Sampled DFT Filter Banks, Sub band Over-Sampled DFT Filter Banks, Volterra Filters, Variable Step-Size (VSS) algorithms, Data Reusing Techniques, Partial Update Adaptive Filtering Techniques and Sub band (SAF) Schemes. These approaches aim to address issues in echo cancellation including the performance with noisy input signals, Time-Varying echo paths and computational complexity. In contrast to these approaches, Sparse Adaptive algorithms have been developed specifically to address the performance of adaptive filters in sparse system identification. In this paper we have discussed some AEC algorithms followed by comparative study with respective to step-size, convergence and performance.