雙邊對話偵測器在迴音消除之性能評比

Abstract

對於迴音消除而言，當雙邊同時對話時，會使得一些採用誤差回授的適應性濾波器如LMS無法追蹤房間的脈衝響應。在本篇論文中，我們將採用修正型相關函數的偵測方法來解決雙邊談話的問題。除此之外，我們延伸一種技術，讓此技術可同時評比雙邊對話偵測器在發生雙邊對話和房間響應突然發生改變的情況，計算此雙邊對話偵測器發生偵測錯誤的機率。另一方面，我們推導使用相關函數的雙邊對話偵測器之偵測錯誤的理論機率，同時還簡化了理論錯誤機率。經過簡化，我們可以看到錯誤機率和假警報機率的關係。我們也提出了一個修正型相關函數可偵測雙邊對話和房間響應突然發生變化。在後面的電腦模擬將會證實我們的推導以及提出的方法。

In the adaptive acoustic echo cancellation, double talk can make the adaptive filter diverge from the optimum. In this thesis, the cross-correlation between the microphone signal and estimate echo is used to judge whether double-talk arises. We also derive the theoretical miss probability as a function of false alarm probability. To distinguish the echo path change from double talk and we also propose a modified cross-correlation double talk detector by microphone energy. We not only develop a way to evaluate DTD algorithm whether double-talk or echo path change arises, but also calculate the miss probability. Computer simulations will validate our derivations and proposed methods.