A log-linear discriminative modeling framework for speech recognition [Elektronische Ressource] / vorgelegt von Georg Heigold

A Log-Linear Discriminative Modeling Framework forSpeech Recognition¨ ¨Von der Fakultat fur Mathematik, Informatik undNaturwissenschaften der Rheinisch-Westfalischen¨ TechnischenHochschule Aachen zur Erlangung des akademischen Grades einesDoktors der Naturwissenschaften genehmigte Dissertationvorgelegt vonDiplom–Physiker Georg HeigoldausLuzern, SchweizBerichter:Professor Dr.–Ing. Hermann NeyPr Dr. Dietrich KlakowTag der mundlichen¨ Prufung:¨ 29. Juni 2010Diese Dissertation ist auf den Internetseiten der Hochschulbibliothek online verfugbar¨ .AcknowledgmentsAt this point, I would like to express my gratitude to all the people who supported andaccompanied me during the progress of this work. In particular, I would like to thank:Prof. Dr.-Ing. Hermann Ney for the opportunity for doing research in this interesting andchallenging area. This work would have not been possible without his continuous interest,advice, and support.Prof. Dr. Dietrich Klakow from Saarland University, Germany, for kindly taking over the taskof the co-referee for this thesis.Dr. rer.-nat. Ralf Schluter¨ for the introduction to speech recognition and discriminative training,and his continuous constructive advice.Patrick Lehnen and Stefan Hahn for the introduction to part-of-speech tagging and theirassistance with the experiments.Thomas Deselaers and Philippe Dreuw for their support with the experiments in handwritingrecognition.