Design of a transform coder for high quality audio signals

The design and implementation of a transform coder which exploits the frequency domain psychoacoustic model of the human auditory system is presented here. The design of an audio coder for consumer applications require tradeoffs to be made in terms of MIPS (million instructions per second), memory and algorithms. Algorithms for each processing stage of the encoder have been designed keeping these concerns in view. In particular, for efficient coding of spectral exponents, a Neural Network based solution is presented.