Audio signal separation device and method thereof

Abstract

When independent component analysis is used to separate each of a plurality of signals mixed in an audio signal, it is possible to solve the permutation problem with high accuracy without utilizing knowledge about the original signal or information about microphone positions or the like. A short-time Fourier transform component generates a spectrogram of the observed signal from the observed signal in the time domain. The signal separating means separates the spectrogram of the observed signal into spectrograms of the individual signals to generate spectrograms of the separated signals. The permutation problem solving means calculates a scale corresponding to a degree of permutation, such as Kullback-Leiblar information calculated by using a multidimensional probability density function, or multidimensional kurtosis, from substantially the entirety of the spectrogram of the separated signal. Based on this scale, the signal at each frequency bin of the spectrogram of the separated signal is swapped between channels to solve the permutation problem.

Description

technical field [0001] The present invention relates to an audio signal separating apparatus and method thereof, which separate a plurality of signals mixed in an audio signal from each other by Independent Component Analysis (ICA). Background technique [0002] In the field of signal processing, attention has been paid to the independent component analysis method in which when multiple original signals are mixed linearly by unknown coefficients, the original signals are separated and restored. If this independent component analysis is applied to audio signals, for example, speech uttered simultaneously by multiple speakers can be observed through multiple microphones, and the observed speech can then be separated for each speaker, or into noise and voice. [0003] refer to figure 1 , the case where individual signals are separated from an audio signal in which a plurality of signals are mixed by using independent component analysis in the time-frequency domain will now be...

AI Technical Summary

Problems solved by technology

Therefore, while the separation is successfully accomplished, the separation destination is often inconsistent

Images