A Stochastic Residual-Based Dominant Component Analysis for Speech Enhancement

Abstract

Noise and sparsity often affect speech signals, leading to serious problems in processing and communication. Speech enhancement is required to improve the quality of the speech signals. This paper introduces a new technique that combines a stochastic approach and dominant component analysis, a variant of principal component analysis for adaptive data analysis. The stochastic approach is a modeling technique that takes into account uncertainty and random fluctuations in the signal. This allows for a more precise estimation of residuals. The proposed method involves estimating residuals using a stochastic approach, which subsequently accumulate into a matrix. Adaptively, we compute the dominant components of the residual matrix. We then use these components to reconstruct clean, enhanced speech. The proposed method aims to forecast sparse data, eliminate noise, and minimally affects crucial data attributes such as energy, covariance, dynamic range, and RMS amplitude.

Keywords

• Adaptive PCA
• affine projection
• noise cancellation
• sparsity removal
• speech enhancement