本論文研究一個麥克風陣列之波束成形系統，差分麥克風陣列擁有體積小、高指向性及與頻率無關的波束圖形等優點，在語音增強的應用中很有潛力，但是差分麥克風陣列有放大白雜訊的問題，抵銷語音增強上的優勢，造成輸出結果不如預期，尤其在高階的差分麥克風陣列越容易發生。 本論文提出譜減法之差分麥克風陣列，在訊號端方面使用譜減法，消除電子原件互相干擾產生的白雜訊，增加輸入訊號的訊雜比；差分麥克風陣列演算法方面，由於傳統時域差分麥克風陣列有設計靈活度低且白雜訊增益低等缺點，因此時頻域差分麥克風陣列架構較多數人使用，此架構的特色將時域訊號透過短時距傅立葉轉到時頻域，透過設計好的濾波器係數，達成保留特定方放的聲音、消除周遭環境音的效果，並且提升白雜訊增益。在時頻域差分麥克風的基礎上，設計一個強健性時頻域差分麥克風陣列，更有效地增加白雜訊增益。;This thesis studies a differential microphone array beamforming system. The differential microphone array has the advantages of small size, high directivity, and frequency-independent beampattern. It has great potential in speech enhancement applications. However, the differential microphone array suffers from a white noise amplification problem, which caused the output not as expected, especially in higher order differential microphone arrays. In this thesis, a differential microphone array with spectral subtraction is proposed. The spectral subtraction method is used in the signal end to eliminate the white noise generated by the mutual interference of electronic components and increase the signal-to-noise ratio of the input signal. The time domain differential microphone array algorithm has low design flexibility and low white noise gain problem. Therefore, the time-frequency domain differential microphone array architecture is often used by many people. The feature of this architecture transforms the time domain signal to the time-frequency domain through the short-time Fourier transform and uses the designed filter coefficient to preserve the sound of a specific direction, eliminate the surrounding ambient sound, and increase the white noise gain. Based on the time-frequency domain differential microphone, a robust time-frequency domain differential microphone array is designed to increase the white noise gain more effectively.
