In this section, two different kernels will be tested on the speech
classification task. These are the Radial Basis Function with radius
, and polynomial function of degree d. The choice of
and d will also be discussed. The SVM formulation does not include
criteria to select a kernel function that gives good generalisation (or
results in a classifier with low expected error bound).
[Smith, 1998] has done extensive work in this area. A generative model
is used to derive a kernel based on the
information from the given training data. The kernel given by this
model gives comparable results compared to the conventional kernel function
used in SVMs.