제안 방법

• The goal of convolutive source separation is to restore original unknown sources from their mixtures where sources are convolved with an unknown multivariate linear time invariant FIR filter which reflects the characteristics of propagation media. In this paper, we trasform the convolutive BSS problem into multiple short-term instantaneous problems in the frequency domain and separating the instantaneous mixtures in every frequency bin in order to tackle this problem [2], [3]. We propose a method based on SEONS (Second Order Nonstationary Source Separation) [1] that was originally developed for blind separation of instantaneous mixtures using nonstationarity.
• Convolutive source separation is a fundamental problem which plays a critical role in cocktail party speech recognition. The goal of convolutive source separation is to restore original unknown sources from their mixtures where sources are convolved with an unknown multivariate linear time invariant FIR filter which reflects the characteristics of propagation media. In this paper, we trasform the convolutive BSS problem into multiple short-term instantaneous problems in the frequency domain and separating the instantaneous mixtures in every frequency bin in order to tackle this problem [2], [3].
• where {)} is the room impulse response between the jth speech source and the zth microphone and n()is the additional sensor noise. The problem is to reconstruct the unknown sources from the sensor data by assuming statistically independent sources without any other prior knowledge about the sources and the mixing process. The frequency domain approach to blind source separation of convolutive mixtures is to transform the problem into multiple short-term instantaneous BSS problems in the frequency domain and the independent component analysis (ICA) is applied to the instantaneous mixtures in every frequency bin.