即時音源可視化分析系統研究

Abstract

本論文以麥克風陣列進行即時音源位置可視化分析系統研究，探討麥克風陣列幾何配置分析和兩種波束形成法(Beamforming)之比較，採用16支麥克風排列成正圓形二維陣列，以波束形成演算法計算二維量測平面的聲音源位置分佈情形，將聲音源分佈圖繪於網路攝影機影像上同時顯示，達成即時音源可視化觀測目的。Delay-And-Sum Beamforming (DAS)為一種低計算複雜度之時域演算法，容易符合即時運算的需求，然而量測解析度較低且易受到環境聲音之干擾；Minimum Variance Beamforming (MV)，具有較高的解析度與高抗雜訊能力，卻存在運算量較大與可用頻寬較小的缺點。本研究討論麥克風陣列之幾何排列方式並比較DAS和MV兩種波束形成演算法應用於即時音源方位(Direction of Arrival；DOA)偵測的效果，基於DAS演算法，設計一種能夠適應環境聲音源位置的聲音源偵測方式，減少聲音源位置誤判的情形，改善傳統DAS演算法量測的效果，最後，探討聲音源位置分析計算法則以提升程式執行效率並降低Beamforming的運算量，實現一個可即時音源可視化分析之系統。

This study develops a microphone array system that integrates two conventional beamforming algorithms and NI-PXI data acquisition (DAQ) system for real-time acoustic source visualizing research. By using 16 channel uniform circular microphone array, it measures sound sources and displays a two-dimensional sound pressure distribution on web-cam video. Delay-And-Sum Beamforming (DAS) is a low computational complexity time-domain algorithm, and easily fits in with a real-time system. However, because of its wide beamwidth and high side-lobe level, a strong interference in the operating environment can easily degrade its performance. Minimum Variance Beamforming (MV) provides higher resolution and adaptive anti-interference ability. Nevertheless, it also has problems with higher computational complexity and narrow-band measurement. This study analyzes microphone array geometric and meanwhile it compares two beamforming algorithms, DAS and MV, for the efficiency in real-time sound source direction of arrival (DOA) detection. Based on DAS, this study proposes an adaptive approach of detecting sound sources to reduce the errors of DOA detection in order to improve the conventional DAS algorithm. The thesis intends to discuss the computation rules of DOA analysis and to improve the beamforming efficiency for practicing a real-time acoustic source visualizing system.