Pre-processing scheme to effectively compensate environment and equipment factors for sound source separation

This work proposes a preprocessing scheme to effectively enhance the performance of sound source separation. In the beginning, time stretched pulses are employed to stimulate an environment with loudspeakers and microphones. The signals recorded from microphones are used to determine physical parameters, including environment and equipment factors, of the system response associated with a real-world environment. According to the determined parameters, the mixture signals are preprocessed by de-convolution to decrease the influence of non-linearity and noise accompanied with such a recording process. The experimental results reveal that our preprocessing scheme can apparently improve the performance at the popularly-used blind source separation algorithms. Therefore, the scheme proposed herein can be beneficial to various multimedia applications demanding sound source separation.