Packet Loss Concealment Using Regularized Modified Linear Prediction through Bone-Conducted Speech

What is it about?

In this paper, we propose a packet loss concealment (PLC) technique for Voice over Internet Protocol (VoIP) using bone-conducted (BC) speech signals by deriving a regularized version of the modified covariance (MC) method. Innately, BC speech has a larger spectral dynamic range which causes ill-conditioned in the conventional PLC techniques. To overcome this problem, we focus on covariancebased linear prediction (LP) methods. In the ground of numerical analysis, we often face an ill-conditioned case that occurs in finding the solution. To deal with this situation, the regularized least squares (RLS) method is employed. Motivated from the RLS concept, we derive the regularized MC (RMC) method for BC speech analysis in the PLC technique. The RMC method has the effect of compressing the spectral dynamic range of the input speech signal, and this effect reduces the numerical ill-condition of LP. Through experiments, we show that the PLC technique based on the RMC method reconstructs more accurate speech than conventional PLC methods for BC speech. The performance of the RMC method is influenced by the setting of the regularization parameter. A way to find the regularization parameter in practice is experimentally derived. The RMC method with such setting provides the best performance in the PLC technique for BC speech

Featured Image

Why is it important?

The RMC method has the effect of compressing the spectral dynamic range of the input speech signal, and this effect reduces the numerical ill-condition of LP. T

Perspectives