This research has developed a novel technique that is based on the fundamental property of background and foreground signals. Background signals are a result of the inferential summation of large number of sources, while the foreground signals are a result of limited number of sources. This makes the statistical properties of the signal very different. Using negative entropy, this article demonstrates that it is possible to obtain the foreground signals from the mixture of foreground and background signals. The technique is based on mixing the noisy recording with a similar known signal and separating the signals using negative entropy based independent component analysis (ICA). The results indicate that the technique is successful in significantly improving the quality of the audio signals.