以FastICA為基礎之時域聲音分離演算法

Abstract

本論文提出一個以快速獨立成分分析法（Fast Independent Component Analysis,FastICA）為基礎的時域聲音訊號分離演算法．藉由模擬觀察到的FastICA特性，發展出一種新的演算法，讓原本無法分離含時間延遲差距（Time-delay difference）訊號的FastICA方法，可以在本論文所提出的演算法架構之下，分離出個別之聲音訊號。本論文所提的演算法，其待處理的兩個混合訊號主要由兩個聲源經由不同的混合系統(mixing system)混合而成．本篇論文將會詳細介紹FastICA原理和及新提出的演算法並提供實現的方法，同時以模擬及實驗驗證此演算法的可行性．

This thesis proposed a time-domain algorithm for audio-signal separation based on FastICA (Fast Independent Component Analysis [b]）. Existing FastICA can not separate signals with time-delay difference directly. In this thesis, two microphones are used to measure the sound sources with time-delay. By pre-processing of the measured signal, the algorithm proposed is able to separate the same signals. FastICA algorithm and the new algorithm will be introduced in detail. The computational procedures to implement the new algorithm are also explained. Simulations and experiments are conducted to prove the effectiveness of this algorithm.