Noise reduction based on spectral entropy in MP3 compressed domain

The compressed format of MP3 is a popular way to store on the personal computers and transmit on internet, but very few algorithms run directly in the compressed domain for noise reduction and it is rather time-consuming to employ traditional methods due to the processing of decompression-noise reduction-compression. In this paper, a novel approach combining with MP3 coding and MDCT coefficients (MDCTs) spectral entropy is proposed to remove directly noise in compressed domain. MDCTs can be extracted from the partial decompression of MP3 audio signal, and then the MDCTs´ entropy is calculated in each granule for noise power spectrum estimation. Experimental results show that the proposed method remove the noise effectively in the compressed domain and improve the efficiency of noise reduction for MP3 audio.