Reliability Driven Linear Estimation of Channel Impaired Speech Parameters

Abstract:
In this paper a linear estimation approach for an improved receiver-side reconstruction of channel impaired speech parameters is presented that is based on the combined use of residual correlations in the coded parameters on the one hand and on reliability information on all received and estimated parameter values on the other hand. In order to tune the estimation to the perceived quality parameter deviations are analysed concerning their perceptual impact and specific mappings for perceptually based reliabilities are derived. Linear adaptive as well as two-dimensional predictors are investigated concerning their potential to capture the spatial and potentially time-varying correlations of vectorial speech parameters with the LSPs as example. Based on these findings two alternative estimation procedures are developed. Training and evaluation is done based on an extensive speech database and an AWGN channel modell. The results show clear improvements, especially for low channel SNR.