The voice production system model

Since LF, a non-interactive parametric model, is chosen to model the glottal source, it means that we are prepared to assume that there is no interaction between the excitation source and the resonant vocal tract. This interaction certainly exists, but due to the high impedance of glottis it is not very significant. The apparent loss in accuracy, induced by such an assumption, is outweighed by the simplifications in speech analysis and modelling. The voice production system can correspondingly be modelled as the source-filter model. The system can be viewed based on their function as consisting of three linearly connected blocks. The glottal source is spectrally shaped by the vocal tract and is radiated by the lip radiation. The model is showed in figure 5.

Figure 5, A model of human voice production system

The lip radiation has very similar spectral properties as a derivative filter and if they are equated the radiation block can be replaced with the filter . Speech is inevitably analysed over small time intervals. This is due to the fact that statistical properties of the speech signal are changing very gradually and in duration of few pitch periods is effectively stationary. In such circumstances the vocal tract and radiation can be considered as linear time-invariant systems. Therefore, they can be interchanged such that the glottal excitation and radiation can be combined to form a derivative glottal wave. On the other hand, the voice production system simplifies to source filter model.

Figure 6, Source Filter Model of voice production system

The Vocal Tract can be modelled as an all pole filter when non-nasal sounds are considered. The resulting system is showed in figure 7. The model should provides a good platform of speech analysis and synthesis systems.

Figure 7, A possible model of human voice production system