T. Murayama and S. Hashimoto (Japan)
Blind source separation, monaural-sound, Modeling, power spectra, Sound number estimation, Matrix equation,
Humans possess an extremely flexible sound analysis ability to separate mixed sounds. Several methods, to real ize automated sound source separation, have been pro posed. The most of them assume certain assumptions on the measurement and the sound sources such as multiple microphones and non-overlapping spectra. It is better to make the assumptions looser. We have also proposed a monaural-sound source separation algorithm based on the time change differences of the power spectra for mostly overlapping sounds. However, the algorithm assumes that the mixed sound is composed of two sounds. In this paper, we propose an improved monaural-sound source separa tion algorithm, which can separate a mixed sound com posed of more than two sounds without the knowledge of the sound source number.
Important Links:
Go Back
