Bangla Hate Speech and Non-Hate Audio Dataset

Published: 27 May 2025| Version 1 | DOI: 10.17632/tgjbpp4z5b.1
Contributors:
,
, Fariha Tasmin Jaigirdar , Sakir Hossain Faruque

Description

This dataset contains 3,000 Bangla audio speech samples divided into three categories: Hate, Non_Hate, and Test nh. The Hate category includes 2,000 speech samples representing various forms of hate-mixed content, such as religious discrimination, gender harassment, political aggression, racism, and ridicule of public figures or TikTok influencers. Each sample is tagged with a severity level: mild, moderate, or severe. The Non_Hate category includes 1,000 speech samples with neutral or non-hateful content. The Test nh folder contains additional non-hate speech samples reserved for testing purposes. The dataset incorporates dialectal variation across Bangla-speaking regions to reflect linguistic and sociocultural diversity. All samples were manually annotated by native Bangla speakers.

Files

Institutions

  • Daffodil International University
  • Deakin University School of Information Technology

Categories

Speech Processing, Speech Recognition, Speech Analysis, Bengali Language, Speech Identification, Freedom of Speech

Licence