The paper describes a new scenario for low bit rate audio compression that combines two classical techniques:
transform coding and sinusoidal coding into a united framework. The main idea is to adaptively decompose the
audio signal into subbands whose central frequencies follow continuously the local instantaneous frequencies of
certain signal components (formants or individual harmonic partials). The content in each subband is encoded in the
baseband after frequency shift towards DC. The technique may be considered either as modified transform coding,
i.e. coding along instantaneous frequencies or as extended sinusoidal coding, i.e. modeling with partial envelopes
that are represented by transform coefficients. In other words, it is a hybrid scheme offering a continuous operating
mode between purely transform and purely sinusoidal compression.