Extended glottal LPS with mixed excitation

A speech-modelling scheme that is capable of reproducing voiced fricatives, plosives, etc. with good accuracy, as well as purely voiced or unvoided sounds is presented. The scheme results in a speech coder with great ability and flexibility in the source modelling. Since it conserves the timing of the source, it can be interpreted as a waveform model, as well as a speech modelling scheme. The scheme requires input speech recorded under good conditions, and an adequate analog preprocessing. Moreover, the analysis time is on the same order as that of the glottal linear predictive coder. These two major limitations make the scheme suitable for applications such as speech storing, etc. From preliminary experiments the bit rate needed in coding the parameters in this model is estimated at 5 kb/s