In this paper, we address the problem of monaural source separation of a mixed signal containing speech and music components. We use Discrete Energy Separation Algorithm (DESA) to estimate frequency-modulating (FM) signal energy. The FM signal energy is used to design a time-varying filter in the timefrequency domain for rejecting the interfering signal. The FM signal energy was chosen due to its good ability to differentiate between speech and music signals using localized information both in time and frequency. We present experimental results which demonstrate the advantages and limitations of the proposed method using synthetic data and real audio signals. © 2010 Elsevier B.V. All rights reserved.