The paper describes requirements, concept
and general structure of audio compression scheme employed
in interactive television iTVP project. A perceptual
coder based on a psychoacoustic model and spatio-temporal
signal decomposition using a hybrid filterbank is proposed.
The coder structure is derived from standard MPEG L3
technique. Some ideas for improving the performance at low
bit rates are introduced. Sinusoidal+noise modeling is proposed
as a means of bandwidth extension. Software implementation
issues regarding real-time performance are discussed
at the end of the paper.