Performance evaluation of an audio perceptual subband coder with dynamic bit allocation

The performance of a perceptual audio coder for high quality low bit rate coding of single notes of orchestra instruments such as pianos, bass drums, snares, triangles, etc., is investigated. The coding system consists of a tree-structured perfect reconstruction filter bank with a dynamic bit allocation based on a backward estimation of the perceptible signal energy. The algorithm is applied to CD-quality elementary instruments sounds. The quality assessment is done on the basis of both objective and subjective measures. The subjective tests were carried out by expert musicians who judged the level of similarity between the original CD sound and the compressed one. Moreover, to investigate the possibility of a real implementation of the algorithm, the complexity cost was taken into account, and the evaluation was carried out for different filter lengths, using 16 bits quantized coefficient filter sets. Results show that transparent coding can be achieved for such signals at 1.6/spl divide/1.7 bit per sample depending on the instruments considered.