Acoustic Modelling for Under-Resourced Languages

Sebastian Stüker
2009
Automatic speech recognition systems have so far been developed only for very few languages out of the 4,000-7,000 existing ones. In this thesis we examine methods to rapidly create acoustic models in new, possibly under-resourced languages, in a time and cost effective manner. For this we examine the use of multilingual models, the application of articulatory features across languages, and the automatic discovery of word-like units in unwritten languages.
doi:10.5445/ir/1000014983 fatcat:7pfagjzw3jbeffmrd6qbnllkym