关键词: 双端会话声回声抵消, 时变功率谱, 代价函数, 归一化步长

Abstract: In order to reduce the high computational complexity load of the double-talk detector in the echo canceller and solve the time-delay problem which leads to the deterioration of the echo canceling performance，a without double-talk detectors based on signal time-varying power spectrum for acoustic echo canceling has been proposed.With the assumption，the near end and the far end speech are independent each other，the far end signal power passing the impulse response of the loudspeaker-enclosure-microphone-system equals the input microphone speech signal power，then using the Least Mean Square error principle for the adaptive filter to cancel the acoustic echo.This method is different from the traditional methods which need to detect the double-talk cases in solving the echo problem during the double-talk periods.Through making the step size normalization，the convergence time of the filter coefficients is reduced.At the same time，the filter becomes more robust.Simulation results show that the proposed algorithm can effectively work in all talking conditions.

Key words: double-talk acoustic echo canceling, time-varying power spectrum, cost function, step size normalization