On the Analysis of Pattern Sequences by Self-Organizing Maps

This thesis is organized in three parts. In the rst part, the Self-Organizing Map algorithm is introduced. The discussion focuses on the analysis of the Self-Organizing Map algorithm. It is shown that the nonlinear nature of the algorithm makes it di cult to analyze the algorithm except in some trivial cases. In the second part the Self-Organizing Map algorithm is applied to several patterns sequence analysis tasks. The rst application is a voice quality analysis system. It is shown that the Self-Organizing Map algorithm can be applied to voice analysis by providing the visualization of certain deviations. The key point in the applicability of Self-Organizing Map algorithm is the topological nature of the mapping; similar voice samples are mapped to nearby locations in the map. The second application is a speech recognition system. Through several experiments it is demonstrated that by collecting some time dependent features and using them in conjunction with the basic Self-Organizing Map algorithm one can improve the speech recognition accuracy considerably. The applications explained in the second part of the thesis were rather straightforward works where the sequential signal itself was transformed for the analysis. In the third part of the thesis it is demonstrated that the Self-Organizing Map algorithm itself could be extended by identifying each Map unit with an arbitrary operator with capabilities for pattern sequence processing. It is shown that the operator maps are applicable for example to speech signal (waveform) categorization. On the Analysis of Pattern Sequences by Self-Organizing Maps Jari Kangas Helsinki University of Technology Laboratory of Computer and Information Science Rakentajanaukio 2 C, SF-02150, FINLAND Thesis for the degree of Doctor of Technology to be presented with due permission for public examination and criticism in the Auditorium F1 of the Helsinki University of Technology on 6th of May 1994, at 12 o'clock noon. i Dedication To my wife P aivi and our little son Miika. ii Acknowledgements The work presented in this thesis has been carried out in the Laboratory of Computer and Information Science in the Faculty of Information Technology of the Helsinki University of Technology. I wish to thank Professor Teuvo Kohonen for initially directing me to work with the SelfOrganizing Map algorithm and for the discussions and the encouragement to continue the research on SOM during the ensuing years. I also wish to thank Professor Kohonen for providing the excellent facilities and support for this research. I would like to thank all the personnel of the Laboratory of Computer and Information Science for creating an inspiring and comfortable atmosphere. The informal discussions with the research group members have always been useful. Professor Teuvo Kohonen and Professor Erkki Oja commented in detail on the manuscript during the nal revisioning which is gratefully acknowledged. Mr. Paul Schurman deserves many thanks for checking the language of the thesis. The nancial support of the Academy of Finland is acknowledged. Espoo, April 15, 1994 Jari Kangas iii