消除空間噪音之研究

Abstract

單麥克風語音加強系統在性能上會受到限制.然而就以語音加強而言雙麥 克風系統表現出較好的結果.一種以雙麥克風做噪音消除的統計方法由 Feder et al. 發展出來.他們用 EM 演算法在頻率領域做運算.在此論文 中, 我們將考慮在串音存在的情況下如何導出 EM演算法在時域消除噪音 的運算方法.此外, 我們將介紹另一種用主動方式做噪音消除的適應性濾 波器 CTRANC.並且提出學習方法用以減低 CTRANC 的收斂時間.在只有噪 音串音存在下, 我們將介紹一種稱為副帶多參考適應性噪音消除器.並且 拿來和 EM 演算法做比較.此論文也評論 filtered--X LMS 演算法和 filtered--U RLMS 對主動噪音控制的適用性. Single-microphone speech enhancement systems have typically shown limited performance. Two-microphone system have shown better results in some context. A statistically based method for noise cancellation in a two microphone system has recently been developed by Feder et al. They use the EM algorithm in the frequency domain. In this thesis, a time domain EM algorithm is derived for removing noise in the presence of crosstalk. And an adaptive filter structure called CTRANC is introduced that actively cancel the noise. And the training method is proposed to reduce its convergence time. Subband multi-reference adaptive noise canceller is also introduced and is compared with the EM algorithm for the condition that only the noise crosstalk exists. The thesis also reviews the application of the filtered-X LMS algorithm and filtered-U RLMS algorithm for active noise control.