On the source-filter model of the vocal tract
Edward P. Neuburg, W. Bauer · 2005
Models of the vocal tract that have a source followed by a linear filter have been of great practical use in Recognition, Compression, Synthesis, etc. Improvements have generally been in the direction of more accurate derivation of parameters. Such refinements now only marginally improve performance, probably because it is linearity, not accuracg of parameters, that is wrong. Inadequacy of the linear model is discussed in this paper, and a simple non-linear model with spring-loaded walls is described.