Audio source separation by source localization with Hilbert spectrum

This paper presents a technique to separate the audio signals from their binaural mixtures based on localizing the sources in the space of interaural differences. Two interaural differences, ITD (interaural time difference) and ILD (interaural level difference), are used as the principal cues to localize and segregate the sources. The Hilbert spectrum is employed to decompose the mixture signals into time-frequency (T-F) space. The sources of the mixtures are considered as disjoint orthogonal in the T-F space. The Hilbert spectrum has a better T-F resolution than Fourier based methods and hence it produces a better disjoint orthogonality of the sources. The separation efficiency, as presented in experimental results using our proposed algorithm, is noticeable in this research area.