Date of Award
Spring 2009
Document Type
Dissertation - Restricted
Degree Name
Doctor of Philosophy (PhD)
Department
Electrical and Computer Engineering
First Advisor
Johnson, Michael T.
Second Advisor
Yaz, Edwin E.
Third Advisor
Heinen, James A.
Abstract
As I am finalizing my dissertation, I am considering the two most valuable experiences in my doctoral process. The first is that of identifying a specific research direction for my Ph.D. study, and the second is that of figuring out how to accomplish it. Generally, a perfect topic would allow a student to accomplish his or her program in less time with better quality, but it is very difficult to find this "right" direction at the early stages of study. The research topics I had been working on originally spanned many different areas in both human speech technologies and bioacoustics, including acoustic enhancement for improving audio quality, acoustic feature extraction at the front-end of recognition systems, and looking at the Lombard effect for investigating the auditory system. Eventually, I settled on acoustic model adaptation as my dissertation topic. I have gained much research experience and knowledge from all these areas. Before I decided my research direction, thorough research in this direction was critical. This taught me what other researchers have done in this area, and which part of this direction was still open. However, one more practical point I often ignored was how those people implemented their metl1ods in terms of experimental work and software programming, so I did not focus until almost the last year in my Ph.D. life on whether I could realistically implement the same experiments as what the people did in their works. I finally realized that it was nothing to be proud of to just understand complicated algorithms, such as the expectation maximization (EM), because the derivation of statistical equations is a fundamental skill to a Ph.D. candidate in electrical engineering. Having an earlier consciousness of programming implementation would have given me a better understanding for the time and effort I needed for this research direction, and help me make a wise decision as to both theory and practice. This Ph.D. work was funded by Dr. Dolittle project, which focuses on development of a broad framework for pattern analysis and classification of animal vocalizations by integrating successful models and ideas from the field of speech processing and recognition into bioacoustics (Johnson et al., 2003). Therefore my work naturally consists both of a theoretical aspect for human speech and a practical aspect for bioacoustic application. Although the field of bioacoustics is challenging due to its multidisciplinary nature, speech technology is the original foundation. I am hopeful that my Ph.D. research will benefit both fields.