Orthogonal decomposition on the backward prediction errors for faster glottal inverse filtering

Glottal inverse filtering (GIF) could be considered as a step forward in human voice modelling through parametrical analysis. The paper begins with a geometrical approach of the GIF problem, directly derived from the linear prediction framework. The most recently proposed algorithm is then reviewed, with a special emphasis on its geometrical interpretation as a decomposition process in a vector space. The authors then introduce a somewhat modified and more efficient algorithm based on orthogonal decomposition on the well-known backward prediction error vectors, and give raw experimental results and further research directions