Multimodal Hate Speech Detection in Dravidian Languages Using Text and Audio Features

The increasing prevalence of hate speech on digital platforms has generated considerable interest in automated methodologies to identify harmful and abusive content. This paper introduces a Multimodal hate speech detection system specifically designed for Dravidian languages, focusing on Malayalam, Tamil, and Telugu. The proposed system integrates both textual and acoustic features to achieve a more robust and nuanced understanding of hate speech. Mel-Frequency Cepstral Coefficients (MFCCs) are extracted from the audio data, while textual data is processed using Term Frequency-Inverse Document Frequency (TF-IDF). The deep learning model employs Long Short-Term Memory (LSTM) networks to process text and a dense neural network for audio features; their outputs are combined in a joint classification layer. The performance of the system is evaluated on datasets in Malayalam, Tamil, and Telugu demonstrating promising results with high accuracy and potential for real-world application. These findings suggest that multimodal approaches enhance the detection of hate speech, particularly in languages characterized by complex linguistic and acoustic features.