Summary: | 碩士 === 國立交通大學 === 電信工程系 === 90 === In the adaptive acoustic echo cancellation, double-talk will make the echo canceller fail to trace the room impulse response. In this thesis, the cross-correlations between i) the microphone input and the estimated echo, and ii) the microphone input and the AEC error, are used to judge whether double-talk arises. We will derive the theoretical cross-correlations, detection thresholds, and the detection delays. For practical nonstationary speech signals, we propose the Variant threshold method to detect the double-talk more efficiently. To distinguish the echo-path change from double-talk, we also propose a Modified-cross-correlation method. Computer simulations will validate our derivations and proposed methods.
|