The MGB-2 Challenge: Arabic Multi-Dialect Broadcast Media Recognition

摘要

This paper describes the Arabic Multi-Genre Broadcast (MGB-2) Challenge forSLT-2016. Unlike last year's English MGB Challenge, which focused onrecognition of diverse TV genres, this year, the challenge has an emphasis onhandling the diversity in dialect in Arabic speech. Audio data comes from 19distinct programmes from the Aljazeera Arabic TV channel between March 2005 andDecember 2015. Programmes are split into three groups: conversations,interviews, and reports. A total of 1,200 hours have been released with lightlysupervised transcriptions for the acoustic modelling. For language modelling,we made available over 110M words crawled from Aljazeera Arabic websiteAljazeera.net for a 10 year duration 2000-2011. Two lexicons have beenprovided, one phoneme based and one grapheme based. Finally, two tasks wereproposed for this year's challenge: standard speech transcription, and wordalignment. This paper describes the task data and evaluation process used inthe MGB challenge, and summarises the results obtained.