Acoustic modelling for under-resourced languages [Elektronische Ressource] / von Sebastian Stüker

Acoustic Modelling forUnder-ResourcedLanguageszur Erlangung des akademischen Grades einesDoktors der Ingenieurwissenschaftender Fakult¨at fur¨ Informatikder Universit¨at Fridericiana zu Karlsruhe (TH)genehmigteDissertationvonSebastian Stuk¨ eraus DetmoldTag der mundlic¨ hen Prufung:¨ 23. Juli 2009Erster Gutachter: Prof. Dr. Alexander WaibelZweiter Gutachter: Prof. Dr. Tanja SchultziiAbstractOver the past decades research in the field of automatic speech recognitionhas lead to systems with a sufficiently high grade of maturity that makes themsuitableforuseinreal-lifeapplications. However,suchrecognitionsystemshavebeen developed only for very few languages. Languages addressed are mainlythose with a large population, a high economic power, or for which a highpolitical interest exists. For the vast majority of the 4,000-7,000 languages inthe world no well performing speech recognition systems exist.Languages are dying at a rapid rate. Linguists estimate that up to 90% oftoday’slanguagesmaygoextinctwithinafewgenerations. Oftenlanguagesdie,because their speakers abandon them in favor of a more wide-spread language,from which they expect more economic or cultural advantages. We believe thattechnology can play a role in stopping this trend, if it were to provide naturallanguage processing technologies, including automatic speech recognition, forall languages in the world.