An Update Algorithm for Frequency-Domain Correlation Model in a Nonlinear Echo Suppressor

This paper proposes a model update algorithm for a nonlinear-echo suppressor (NLES) based on a frequency-domain correlation between linear echo replica and residual nonlinear echo. The regression coefficient of the correlation obtained by the proposed algorithm is minimum tracking of the amplitude ratio of echo replica and residual echo in the frequency domain. The influence by the near-end signal is reduced by the minimum tracking, resulting in stable regression coefficients of the frequency-domain correlation. Simulation results demonstrate that the NLES with the proposed update algorithm achieves comparable echo suppression performance to the NLES with fixed regression coefficients tuned in advance, and that it can track changes of echo-path nonlinearity.