Transform coding of audio signals using correlation between successive transform blocks

The authors propose a transform coding algorithm for bit rate reduction of high-quality sound. The algorithm provides for long-term stationarity in the transform coder while keeping fixed blocklength. Side information is drastically reduced, and predictive coding of DFT (discrete Fourier transform) coefficients is used to remove the interblock redundancy. Perceptual properties are incorporated in the bit-allocation procedure to achieve spectral noise shaping. The complete algorithm is described, and results of coding at 96 kb/s for a monophonic 15-kHz audio signal are discussed