Wouldnt it be easier to fine tune a model to convert the Brazilian Portuguese corpus into European Portuguese and then use that corpus?