Perceptual considerations in a low bit rate sinusoidal vocoder

An approach to coding the parameters of a harmonic sinusoidal model which incorporates error spectrum shaping in order to improve the subjective quality of the speech coded at low bit rates is presented. It is shown that the sinusoidal model formulation is very well suited to representing speech signals and provides knowledge of pertinent time-varying characteristics of speech, such as pitch and short-time spectral information, which is useful for speech coding. The model also has a particularly simple form which lends itself easily to analysis and includes an envelope signal which separately models syllabic volume changes, enhancing the performance of the model. The analysis procedure presented provides greater accuracy than other techniques, resulting in higher quality synthetic and coded speech. In addition, as more components are added, the synthetic speech signal is guaranteed to converge to the original speech signal. Using perceptual factors in coding the parameters of this model yields considerable improvement in the overall subjective performance of the coder