Reconstruction of missing speech frames using sub-band excitation

A new reconstruction method for frame erasures in speech transmission is presented which is based on parameterization of the speech signal by means of linear prediction (LPC) and voicing analysis. The problem of generating partially voiced substitute speech signals is solved by performing separate voicing decisions in sub-bands. The method yields considerable improvements compared with silence substitution for frame erasure ratios of up to 10% or even 20%. The combination of the reconstruction method with adaptive speech coders showed virtually the same good results for forward adaptation, whereas a higher degradation is caused by backward-adaptive coders.