Phase-aware Audio Inpainting Based on Instantaneous Frequency

In this paper, we propose a phase-aware audio inpainting method and its variations. The aim of audio inpainting is to restore some gaps of audio signals. Among various methods, sparsity-based methods have been studied, especially for restoring gaps up to 50 ms. However, they usually focus only on the magnitude information even though the phase information is also available. In the proposed method, we utilize a phase-aware prior to induce smoothness of phase. The variations of the basic algorithm are also obtained by applying some updating schemes. The numerical results show effectiveness of the proposed methods, especially for longer gaps.