Residual echo is the echo that remains after a linear echo canceller has done its best — the part it cannot model, arising from nonlinear loudspeakers (cheap drivers that distort), clock drift between capture and playback, or echo paths that change faster than the filter adapts. Because the linear filter can't subtract what it can't predict, a separate residual-echo suppressor — often integrated with the noise-suppression stage — applies nonlinear processing to attenuate whatever leaks through. This two-stage design, linear cancellation plus residual suppression, is standard in modern systems like AEC3, and getting the residual stage right is what stops the faint, intermittent echo that survives on speakerphones and Bluetooth devices.
