利用訊號特徵及麥克風陣列

Abstract

本論文針對室內訊噪比(SNR)很低的吵雜環境提出辨識出雜訊和聲源的聲音監控系統。在傳統上單顆麥克風在訊噪比很低的情況下，無法辨識出雜訊和聲源。在此藉由麥克風陣列訊號擷取系統以擷取多通道聲音資訊，利用空間濾波器來抑制干擾源的影響可以提高訊噪比(SNR)降低偵測的錯誤率。並利用訊號特徵與高斯混合模型(Gaussian Mixture Model)的方法去建立聲音監控的背景模型，在此用期望值最大演算法(EM Algorithm)去估計模型參數。最後結合背景模型中聲場特徵的統計資訊，做背景聲音和聲源的判定。本研究以USB1.1介面、8通道麥克風陣列訊號處理實驗平台進行研究，以此實驗平台錄製麥克風陣列語音樣本並供相關研究。

The purpose of this thesis is to construct an audio surveillance system to recognize the interference sources and sound sources in a low signal-noise ratio (SNR) noisy environment. Traditionally, single microphone can not recognize the noise and sound sources in a low signal-noise ratio (SNR). First, the time delay between microphones are estimated from the multiple-channel sound data acquired by the digital microphone array acquisition system. Second, we suppress the effect of the interference sources by spatial filter to promote signal-noise ratio (SNR) and to reduce the false detection ratio. Second, we create the background model of surveillance system by using the signal characteristic and the Gaussian Mixture Modeling method, and then to estimate the model parameters by utilizing the EM algorithm. Finally, we can recognize the sound sources from the interference sources through the statistic information of sound field characteristic. A microphone array of 8 microphones with USB 1.1 interface was made for the implementation platform.