摘要 |
A device and a method for separating audio signals are provided to solve a permutation problem at high precision without using knowledge related to original signals and information related to a position of a microphone in separating audio signals by using independent component analysis. A short-time Fourier transform unit(12) transforms an observation signal of a time domain into a signal of a time frequency domain to generate spectrogram of the observation signal. A separator(13) generates spectrogram of a separated signal from the spectrogram of the observation signal. A permutation problem solving unit(15) solves a permutation problem of the spectrogram of the separated signal. The permutation problem solving unit calculates a measure corresponding to degree of permutation, from almost whole spectrogram of the separated signal, and solves the permutation problem by exchanging a signal of each frequency bin of the spectrogram of the separated signal between channels according to the calculated measure.
|