獨立成份分析法於真實環境中聲音訊號分離之探討

A Study of Field Audio Signal Separation by Using Independent Component Analysis

蔡政哲(Cheng-Tse Tsai)

指導教授 ： 吳炤民

國立中央大學/資訊電機學院/電機工程學系/碩士(2006年)

摘要

中文摘要 本論文的主要工作在透過真實環境中的混合聲音，特別是語音，驗証獨立成份分析法(Independent Component Analysis, ICA)能夠擷取在兩獨立聲源中具有時間延遲之聲源的特性，目的是希望藉由ICA對聲音訊號分離(audio signal separation)的結果來了解ICA在實際應用上的限制與特性。就理論而言，在統計獨立的假設前提下，ICA可以分離出即時混合(instantaneous mixing)的聲音訊號。然而，環境中的聲音訊號有時間延遲(time delay)、迴響(reverberation)的情況，使得觀察到的訊號為一迴旋混合(convolutive mixing)，造成ICA無法正確地分離出獨立聲源。由我們的理論推導發現，在只考慮兩個聲源及兩支麥克風的情況下，若其中一個聲源在兩支麥克風之間有時間延遲，而另一個聲源沒有，則具有時間延遲的聲源可以被ICA分離出來。最後透過在真實環境中的聲音錄製，實驗証實其結果與理論似乎相吻合。

關鍵字

聲音訊號分離 ； 獨立成份分析法 ； 迴旋混合 ； 迴響

並列摘要

Abstract
The main purpose of this study is to understand the characteristics of independent component analysis (ICA) method and its limits in application of audio signal separation. Theoretically, ICA may separate the instantaneous mixing audio signal under the assumption of mutual statistical independence. However, under the effects of time delay and reverberation, the field-recorded audio signals could be convolutively mixed and could not be correctly separated by the ICA. From our theoretical derivation, the audio source with time delay could be separated by the ICA from the mixed signals that were recorded with two microphones under the condition of one audio source with time delay and the other without. In this study, our theoretical derivation was further tested with field audio signal recordings and the results seem to fit theoretical derivation well.

並列關鍵字

Reverberation ； Convolutive Mixing ； Independent Component Analysis ； Audio Signal Separation

參考文獻

Arora, J. S., “Introduction to Optimum Design”, McGraw Hill, New York, 1989.

Bell, A. J., Sejnowski, T. J., “An information maximisation approach to blind separation and blind deconvolution”, Neural Computation, Vol.7, pp.1129-1159, 1995.

Bell, A. J., Sejnowski, T. J., “Learning the higher-order structure of a natural sound”, Network: Computation in Neural Systems, 7:261-266, 1996.

Component Analysis and Blind Signal Separation (ICA2003), Nara,

Japan, April, 2003.