A MDCT based auditory spectrum model and its application in compressed domain

In this paper, we construct a MDCT-based auditory spectrum model and investigate the calculation of auditory spectrum in audio classification applications. It is focus on the compressed-audio. The MDCT coefficient maintains some of the acoustic features itself, and we build the auditory spectrum model based on the coefficient to simulate the human ear characteristics better. The proposed model is composed by three steps: outer and middle ear modelling; critical band decomposition and the spectrum spreading. Filter is the key point in the processes, and its band architecture would realize cochlea implants or auditory processors of biorealism. To assess the performance of the proposed MDCT-based auditory spectrum model, a range of various speech/music classification tasks is implemented wherein a support vector machine (SVM) algorithm is used as the classifiers. Features used for classification include MFSCs derived from the improvement of conventional MFCCs, as well as other features of the energy category. Compared to the conventional features, the features derived from the proposed MDCT-based auditory spectrum model show higher robust performance in noisy test. Experiments also indicate that, using the new features, the performance of the proposed MDCT-based auditory spectrum is slightly better than that of the original auditory spectrum.