Echo removal is applied to the received audio signal using only the second model estimate of the echo.