Audio Inpainting Using Autoregressive Modeling of CWT Coefficients with Matched Wavelets

Audio inpainting is the task of reconstructing missing or degraded segments in audio signals. It plays a crucial role in signal restoration and transmission applications. This paper introduces a novel approach that integrates matched wavelet-based Continuous Wavelet Transform (CWT) with Autoregressive (AR) modeling for effective audio inpainting. The matched wavelet is optimized using a differential evolution algorithm to align with the signal's unique characteristics ensuring a robust time-frequency representation. The AR model leverages the quasi-periodic nature of audio signals to predict missing coefficients in the wavelet domain, employing surrounding coefficients as training data. Experimental results show that the reconstructed signals achieve high cross-correlation coefficients with the original signals reflecting accurate reconstruction. The Mean Opinion Score (MOS) evaluations indicate that the perceptual quality of the reconstructed audio is maintained across various gap durations. These findings highlight the effectiveness of the proposed approach and its potential applications in audio restoration and signal processing.