The fractal characterisation of phonetic elements of human speech

The use of fractal techniques and fractal dimensions as a means of speech characterisation and speech recognition is a relatively new concept and as such very few papers have addressed the possibilities of its use and associated advantages and disadvantages over conventional methods. This thesis demonstrates that fractal techniques can effectively be used as a method of broad recognition of phonetic elements in human speech. Three distinct fractal methods have been used to associate fractal dimensions with speech: the Box Counting method, the Divider or Richardson method and the Minkowski-Bouligand disc method. Speech has been recorded by myself and another male and female speaker to provide a database of phonetic recordings that could be experimented on. The three fractal techniques were emulated by means of software programs written in a high level language.