Audio denoising in MP3 compressed domain

In this paper, we present a novel audio denoising algorithm for MP3 (MPEG1-layer3) compressed audio, based on Modified Discrete Cosine Transform (MDCT). The proposed algorithm works directly on MP3 frames. Firstly, we get MDCT coefficients from MP3 audio semi-decoder. And then, the normal inverse Gaussian (NIG) probability density function is performed to modelling the statistical distribution of MDCT coefficient, based on its sparse features. In the denoising process, a maximum a posterior (MAP) estimator based on NIG density is exploited to remove noise from MP3 audio. The experimental results show that the proposed algorithm can remove noise directly from noisy MP3 audio, and the signal-to-noise ratio (SNR) of the compressed audio is improved obviously.