Abstract: Methods and systems pertaining to preventing the corruption of filter coefficients due to near end speech in an echo canceller that uses an adaptive filter to create a local replica of an echo signal are provided. In accordance with a preferred method, a far end signal is received and the presence of near end speech is indicated. A filter coefficient vector is updated, and copies of the filter coefficient vector are periodically stored. A stored copy of the filter coefficient vector is recovered for use in the echo canceller. Further, an improved echo canceller that prevents the corruption of filter coefficients due to near end speech, having a near end speech indicator, a first buffer, a filter, and a second buffer connected to the filter, is provided. The first buffer receives a far end signal. The filter is responsive to the first buffer and the near end speech indicator.

Abstract: Methods and systems pertaining to estimating and compensating for a round-trip or flat delay portion of an echo channel characteristic in an echo canceller are provided. According to a particular method, a filter coefficient vector of an adaptive filter that models the echo channel characteristic is updated. An instant flat delay is estimated by analyzing the filter coefficient vector after each update, and periodic averages of the instant flat delay estimate are performed. A final flat delay estimate average is determined when the flat delay estimate average has converged. The length of the adaptive filter is adjusted in response to the final flat delay estimate average. Further, an improved echo canceller apparatus that compensates for a flat delay portion of an echo channel characteristic is provided. The improved echo canceller includes an adaptive filter, a position detector, a periodic averager, and a level detector. The adaptive filter has an initial length N.