I. Introduction
Blind source separation (BSS) aims to estimate source signals from signals observed by one or more microphones without any prior information about mixing systems. It is a fundamental technique for audio applications such as automatic speech recognition (ASR) and music transcription since separated signals are required as input.