Abstract: A multi-block frequency-domain adaptive filter in an acoustic echo canceller as applied to a full-duplex speakerphone includes convolving a received reference signal with a linear room model to generate an echo free signal for transmission which, in turn, is used to adapt the coefficients of the room model. The frequency-domain estimated echo response is transformed into the time-domain to generate an approximated time-domain echo signal which is subtracted from a microphone signal containing a desired intelligence signal and the actual echo to produce an outgoing signal. The outgoing signal serves as an error signal used to adapt the filter. In the preferred embodiment, a weighting means adjusts the frequency-domain room coefficients during the convolution operation. In another embodiment, a weighting means adjusts the frequency-domain room coefficients by adjusting the correlation terms during adaptation of the frequency-domain room coefficients.