A novel hybrid DWPT and MDCT based coding technique for sounds of musical instruments

In this paper, a new hybrid multi-stage musical instrument sound signal compression method, based on DWPT and MDCT using efficient psychoacoustic models is proposed. The primary objective is to perform lossy and perceptually transparent compression on the sounds of various Indian musical instruments. Firstly, the original sound signal is decomposed into wavelet packets using an optimal wavelet basis. Wavelet packets are then run through psychoacoustic model in wavelet domain to determine auditory masking level for thresholding, which in turn is used to perform thresholding of wavelet coefficients in respective sub-bands. Further, audio signal coefficients are partitioned into frames which overlap in such a way that each block gets 512 samples and windowed using a hanning window with 1/16 frame overlap. Secondly, MDCT is applied to each block to de-correlate the spectral information. Removal of spectral redundancy is achieved by compressing the subordinate components more than the dominant components. The resulting signal is quantized with variable number of bits, which are determined based on the results of the psychoacoustic model in FFT domain. This technique provides an efficient way to exploit key strengths of both DWPT and MDCT.