A comparison of audio compression transforms

This paper provides a subjective quality analysis of transforms used in audio compression algorithms for a class of music signals. An analysis with the discrete wavelet packet transform (DWPT) compares performances of an eight-level non-uniform critical-band frequency split versus a five-level uniform subband frequency split used in MPEG. Another analysis compares the performances of the DWPT versus the modified discrete cosine transform (MDCT). The quality assessor at each level of compression was based on a listening test (34 subjects). Results show that the critical-band split provides significantly better quality than the even-band split for sounds with strong low frequency content, and that the DWPT outperforms the MDCT overall, but with significant improvement for non-tonal sounds.