Abstract: According to the features of speech signals, the paper proposes a robust speech transfer path modeling system. After adding a low-level broadband random training signal to eliminate the modeling mismatch, it removes the speech signals in the system output by using an adaptive prediction filter, and it also modifies the distortion of the training signals due to the APF by using a compensation filter. The computer simulations confirm that the novel modeling system can be widely applied to echo cancellation systems and speech enhancement systems.

Key words: adaptive filter, speech enhancement, adaptive modeling, echo cancellation