Abstract: The detection of echo is enhanced using a coarse near-end speech detector at a Y-side of an echo canceler and using a more accurate near-end speech detector at the E-side of the echo canceler, such that the latter detector uses a representation of far-end speech and a threshold value to determine if the magnitude of the echo canceler output is larger than that of an echo signal only. If it is, and there is an absence of an indication of the presence of near-end speech from the coarse detector, then the more accurate speech detector adjusts the value of the threshold in a first direction, otherwise it adjusts the threshold in a second direction, to more "finely tone" such accuracy.