音景訊號分離與降噪之整合型系統研製

在噪音消除技術的發展中，許多研究曾提出各式演算法，致力於降低環境中的噪音成分，然而多數研究僅侷限於降低特定頻帶之噪音，此類方法對於複雜且多變的環境通常無法有顯著的效果。另外，亦有主張以適應性學習法在多變的環境中達到降低噪音之研究，但同時也會將一些重要的聲音訊息一併削弱，並非我們所樂見的。本論文利用環境中之音源訊號在空間中往往互為線性獨立之物理特性，以獨立成份分析法(Independent Component Analysis, ICA)中的最大資訊量法(Infomax ICA)，在各音源皆為未知的狀態下，僅憑藉麥克風擷取到之混合音頻訊號，將各自獨立的音源訊號由音景中萃取出來，達到盲訊號源分離(Blind Source Separation, BSS)。分離後之音源訊號可透過本研究開發出之介面做試聽與選取，並結合空間域之主動式降噪(Active Noise Control, ANC)，以降低被選取之音源訊號在三維空間中之能量強度。本研究除了可以針對特定音頻訊號進行削減外，更進一步地可將欲留下之音源訊號重新組成我們所期待的音景，此為目前已知文獻中未曾涉及之研究。　　本論文除了利用多元之音源訊號於MATLAB中，以離線方式進行模擬，並具有優良響應與高度可行性外，亦將演算法拓展至嵌入式系統中，能夠即時地自音景中分離出獨立之音源訊號，除了可以降低所選取的訊號源(通常為噪音)於指定區域中之能量強度，並能針對好聽的聲音源進行重組。

關鍵字

盲訊號源分離；獨立成份分析；主動式降躁

並列摘要

In the development of noise-canceling technology, a variety of algorithms have been proposed by different research studies that strived to minimize noise components in the surrounding. However, the majority of research studies are limited to reducing specific noise frequencies. Methods belonging to this category have no significant effect on real environments, which are complex and variable. Additionally, some other research studies propose using adaptive learning to achieve noise reduction. However, this method also tends to eliminate important sound data, which isn't something we look forward to. This study, according the usually linear independent physical characteristic of sound source signals in the environment, makes use of Infomax ICA, an Independent Component Analysis (ICA) method. In the condition that every sound source is unknown, Infomax ICA relies only on mixed audio signals recorded by a microphone to extract each independent sound source signal from the soundscape, thus achieving Blind Source Separation (BSS). After separation, sound source can be checked and chosen via the interface developed in this study. At the same time, integrating Active Noise Control (ANC) of spatial domain, the system minimizes energy intensity of the chosen sound source signals in three-dimensional space. Not only can the system mute specific audio signals here, it can furthermore take the kept sound source signals and reorganize them to form a desired soundscape. This is a research area not yet explored by available references. Besides using diversified sound source signals in MATLAB, offline simulation and possessing excellent response and a high degree of feasibility, this study also expanded by developing an algorithm into an embedded system for instantaneous separation of independent sound source signals from a soundscape. In addition to reducing the energy intensity of selected source (usually noise) in a specified area, the system reorganizes the preferred sound sources.

並列關鍵字

blind source separation ； independent component analysis ； automatic noise canceling

參考文獻

[1] 勞工安全衛生研究所，勞工聽力危害預防手冊，臺北市，行政院勞委會勞工安全衛生研究所， 2008.

[5] B. Widrow and S. D. Stearns, Adaptive Signal Processing, Cliffs, NJ: Prentice- Hall, 1985.

[6] S. M. Kuo and D. R. Morgan, “Active Noise Control: A tutorial review,” Proceedings of the IEEE, vol. 87, no. 6, Jun. 1999, pp. 943-973.

[7] J. V. Stone, “Independent Component Analysis: an introduction,” Trends in Cognitive Sciences, vol. 6, no. 2, 2002, pp. 59-64.

[8] A. J. Bell and T. J. Sejnowski, “An information-maximization approach to blind separation and blind deconvolution,” Neural Computation, vol. 7, no. 6, 1995, pp. 11129-1159.

國際替代計量

音景訊號分離與降噪之整合型系統研製

全文下載

主題瀏覽