Performance Evaluation of Acoustic Echo and Noise Canceller with Variable-Step-Size Shared-Error NLMS Algorithm under Double-Talk Conditions

In this paper, we present the evaluation of the acoustic echo and noise reduction performance of the variable-step-size shared-error normalized least mean square (VSS-SENLMS) algorithm under double-talk conditions. The VSS-SENLMS algorithm was developed for the acoustic echo and noise canceller (AENC) to simultaneously optimize both the acoustic echo canceller and the adaptive noise canceller by variable step size parameters. Our previous study showed that the VSS-SENLMS-based AENC achieved reductions of the acoustic echo and background noise under the single-talk condition with nonstationary background noise. Hence, we evaluate the acoustic echo and noise reduction performance of the VSS-SENLMS-based AENC under double-talk conditions. By computer simulation, this system is found to diverge under the double-talk condition owing to the poor estimation accuracy of variances for variable step size parameters. On the other hand, this system maintains the acoustic echo and noise reduction performance without the double-talk detector when the variances used for the variable step size parameters are accurately estimated.