Audio conferencing systems typically include an echo canceller to remove echo from the loudspeaker that is picked up by the microphone. The echo canceller includes a linear echo canceller for removing linear echos, and a residual echo suppressor for suppressing residual non-linear echos. The residual echo suppressor functions based upon a characterization of the inter-modulation products generated by the audio system, e.g., provided as a distortion map generated using a single-tone sweep at different volumes on the loudspeaker and measuring the distortion at the microphone. Extending this approach to multi-tone input signals has prohibitive memory requirements.