Bangla Hate Speech and Non-Hate Audio Dataset

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.