Separation of Sound Sources Propagated in the Same Direction(Blind Source Separation, <Special Section>Multi-channel Acoustic Signal Processing)

概要

論文の詳細を見る
This paper describes a method for separating a target sound from other noise arriving in a single direction when the target cannot, therefore, be separated by directivity control. Microphones are arranged in a line toward the sources to form null sensitivity points at given distances from the microphones. The null points exclude non-target sound sources on the basis of weighting coefficients for microphone outputs determined by blind source separation. The separation problem is thereby simplified to instantaneous separation by adjustment of the time-delays for microphone outputs. The system uses a direct (i.e. non-iterative) algorithm for blind separation based on second-order statistics, assuming that all sources are non-stationary signals. Simulations show that the 2-microphone system can separate a target sound with separability of more than 40dB for the 2-source problem, and 25dB for the 3-source problem when the other sources are adjacent.
社団法人電子情報通信学会の論文
2005-07-01