ADIMA: Abuse Detection In Multilingual Audio

TL;DR

This paper introduces ADIMA, a comprehensive multilingual audio dataset for abusive content detection in Indic languages, addressing challenges in speech recognition and language diversity to improve audio-based moderation.

Contribution

The paper presents ADIMA, a new large-scale, ethically sourced, expert-annotated multilingual audio dataset specifically designed for profanity detection in Indic languages.

Findings

ADIMA includes 11,775 audio samples in 10 Indic languages.

Experiments demonstrate the dataset's effectiveness in monolingual and cross-lingual settings.

The dataset facilitates future research in audio-based content moderation for underrepresented languages.

Abstract

Abusive content detection in spoken text can be addressed by performing Automatic Speech Recognition (ASR) and leveraging advancements in natural language processing. However, ASR models introduce latency and often perform sub-optimally for profane words as they are underrepresented in training corpora and not spoken clearly or completely. Exploration of this problem entirely in the audio domain has largely been limited by the lack of audio datasets. Building on these challenges, we propose ADIMA, a novel, linguistically diverse, ethically sourced, expert annotated and well-balanced multilingual profanity detection audio dataset comprising of 11,775 audio samples in 10 Indic languages spanning 65 hours and spoken by 6,446 unique users. Through quantitative experiments across monolingual and cross-lingual zero-shot settings, we take the first step in democratizing audio based content moderation in Indic languages and set forth our dataset to pave future work.