A Rule Based Pronunciation Generator and Regional Accent Databank for Portuguese

One of the major obstacles in deploying spoken language technologies (SLTs) in the developing world is a lack of key linguistic resources – e.g. electronic dictionaries, phonetically aligned corpora, pronunciation lexicons, etc. – that describe the non-dominant varieties spoken in such countries and regions. In this paper, we describe the work of the LUPo (Portuguese Unisyn Lexicon) project to model standard and non-standard varieties of spoken Portuguese from around the globe, and: (1) deliver a free, open-source tool for the automatic generation of accent-specific pronunciation lexica within the existing online lexical knowledge base, the Portal da Língua Portuguesa; and (2) provide the research and speech technology communities with a free, online, searchable database, the Portuguese RADbank, dedicated to the description of regional varieties of spoken Portuguese. Both resources are presented as bases for adapting SLTs to regional varieties spoken in the Luso-African and Luso-Asian world, as well as to non-standard varieties of Brazilian and European Portuguese.