Preprint
Speech Recognition Challenge in the Wild: Arabic MGB-3
Ahmed Ali et al., arXiv, 2017
Read the originalRead on October 17, 2026
- What it is
- The 2017 paper describing the MGB-3 challenge, which tested Arabic speech recognition on Egyptian YouTube videos rather than broadcast news. It lists the genres and the size of the test data.
- How much weight it carries
- Describes a research challenge dataset. Its 16 hours of audio are small and were drawn from a few genres, so it shows the dialect problem, not the error rate of any product.
Facts taken from this source
- The MGB-3 Arabic challenge data totals 16 hours of video, split evenly across seven genres.
“A total of 16 hours of videos, split evenly across the different genres” (Abstract)
- The 2017 MGB-3 Arabic challenge emphasized dialectal Arabic and drew its data from Egyptian YouTube videos.
“MGB-3 emphasises dialectal Arabic using a multi-genre collection of Egyptian YouTube videos” (Abstract)