This paper presents a new acoustic echo canceller combined with the linear predictive speech coders

which includes an adaptive FIR (Finite Impulse Response) filter employing the variable step-size de-correlation NLMS (Normalized Least Mean Square) algorithm and a double-talk detector based on the variance of echo-path mismatch.The simulation results conducted by Matlab demonstrate that the proposed algorithms have better echo cancellation performances compared with the echo cancellation algorithms proposed by Gordy.It detects more correctly in both double-talk and echo-path change

converges faster and achieves an average 15dB higher ERLE and 10dB lower misalignment during convergence.