Transcription of Multi-variety Portuguese Media Contents

Current automatic transcription technology applied to media contents is an important medium that not only allows generating subtitles, but also enables data search and retrieval capabilities over multimedia streams. Among others, one of the most important challenges that transcription systems have to deal with is speaker accent variability. In this work we study the importance of accent variability for three broad varieties of Portuguese: African Portuguese, Brazilian Portuguese and European Portuguese. Then, we propose a multi-variety transcription system based on the combination of variety identification followed by specific variety-dependent transcription systems.