An improved residual-domain phase/amplitude model for sinusoidal coding of speech at very low bit rates: a variable rate scheme

An improved harmonic sinusoidal model is presented, where the underlying sine wave amplitudes and phases are efficiently represented using a combination of linear prediction, linear phase alignment, all-pass filtering, and spectral sampling in the residual-domain. The analysis and synthesis systems are introduced and the derivation and encoding of each model parameter is discussed. Performance analysis on a large database indicates effective modeling of the sinusoidal parameters. A variable-rate sinusoidal coder operating at an average bit rate of 1.75 kbps, based on the proposed model, has been developed, yielding reproduced speech of good quality, intelligibility, and naturalness. The proposed model may find applications in low bit rate speech coding in high capacity wireless communication systems.