Network of Data Centres (NetDC): BNSC - An Arabic Broadcast News Speech Corpus

Broadcast news is a very rich source of Language Resources that has been exploited to develop and assess a large set of Human Language Technologies. Some examples include systems to: automatically produce text transcriptions of spoken data; identify the language of a text; translate a text from one language to another; identify topics in the news and retrieve all stories discussing a target topic; retrieve stories directly from the broadcast audio and extract summaries of the content of news stories. BNSC is a broadcast news speech corpus developed in the framework of the European-funded project Network of Data Centres (NetDC). The corpus contains more than 20 hours of Arabic news recordings in modern standard Arabic. The news was recorded over a period of 3 months and were transcribed in Arabic script. The project was done in corporation with the LDC (Linguistic Data Consortium), which has produced a similar corpus of its Voice of America Arabic in the United States. This paper presents the BNSC corpus production from data collection to final product.