"RoU-AudioTox: Roman Urdu Audio-Text Dataset for Offensive Speech Detection and Transcription"

"Since online communications are growing, those who speak Roman Urdu experience more problems with offensive statements and online bullying. English gets a lot of attention from researchers and is supported with many specialized datasets, but Roman Urdu does not, despite being used on social media for Urdu communication. Our solution to bridge this gap is a Roman Urdu offensive multimodal dataset that has been labeled as Abusive, Threatening, Anti-National or Neutral. Our dataset includes recorded audio from a variety of speakers, along with well-labeled transcripts. This database allows researchers to test models for turning offensive speech into text and for detecting offensive content, mainly for low-resource language Roman Urdu. As a result, many of us in the community benefit from being able to fight cyberbullying and communicate safely and easily online."