ADIMA
X
A multilingual, expertly annotated and well-balanced dataset for abuse detection in audio chatrooms.
ADIMA comprises of 11,775 audio utterances (~5-60 seconds) in 10 Indic languages spanning 65 hours and spoken by 6,446 unique users.
10 Indic Languages
ADIMA has been sourced from real life conversations and can be used for developing content moderation algorithms for enabling safe and healthy interactions.
Publications
Please cite the following if you make use of the dataset.
ADIMA: Abuse Detection In Multilingual Audio
ICASSP, 2022
Vikram Gupta, Rini Sharon, Ramit Sawhney, Debdoot Mukherjee
Multilingual and Multimodal Abuse Detection
INTERSPEECH, 2022
Rini Sharon, Heet Shah, Debdoot Mukherjee, Vikram Gupta
Team
License
ADIMA dataset is available for only reseach purposes. Any commercial use of the dataset is strictly prohibited.