Speech Enhancement using Kernel and Normalized Kernel Affine Projection Algorithm

TL;DR

This paper explores the use of Kernel and Normalized Kernel Affine Projection Algorithms in the Reproducing Kernel Hilbert Space to improve speech signal enhancement by effectively removing background noise, outperforming traditional adaptive filters.

Contribution

It introduces the application of Kernel Affine Projection Algorithms in RKHS for speech enhancement, demonstrating improved noise removal over conventional methods.

Findings

Enhanced signal-to-noise ratio in noisy speech

Kernel methods outperform traditional adaptive filters

Effective noise removal in stationary and non-stationary noise conditions

Abstract

The goal of this paper is to investigate the speech signal enhancement using Kernel Affine Projection Algorithm (KAPA) and Normalized KAPA. The removal of background noise is very important in many applications like speech recognition, telephone conversations, hearing aids, forensic, etc. Kernel adaptive filters shown good performance for removal of noise. If the evaluation of background noise is more slowly than the speech, i.e., noise signal is more stationary than the speech, we can easily estimate the noise during the pauses in speech. Otherwise it is more difficult to estimate the noise which results in degradation of speech. In order to improve the quality and intelligibility of speech, unlike time and frequency domains, we can process the signal in new domain like Reproducing Kernel Hilbert Space (RKHS) for high dimensional to yield more powerful nonlinear extensions. For experiments, we have used the database of noisy speech corpus (NOIZEUS). From the results, we observed the removal noise in RKHS has great performance in signal to noise ratio values in comparison with conventional adaptive filters.