Human Language Technologies (HLT) - DictionaryMaker

The availability of language resources is key when developing any speech or language technology in a new language. These resources are expensive to generate and require significant expertise, and can become unaffordable when repeated a number of times for a number of languages. The DictionaryMaker project demonstrates how bootstrapping can be used to develop an electronic pronunciation dictionary – one of the key resources when developing speech recognition or speech synthesis systems – in a fraction of the time typically required, by combining interaction with a first language speaker of the specific language with a machine learning approach. The tool guides the user through the dictionary creation process, which can be completed in a fraction of the time it usually takes to create a pronunciation dictionary. Along with the dictionary, a related set of grapheme-to-phoneme rules is created automatically.

The DictionaryMaker was used to create the pronunciation rules for the isiZulu Text-to-Speech System and the Sepedi Speech Recognition System

Publications