Audio-based Indonesia Toxic Language Classification using Recurrent Neural Network, Speech Recognition and Natural Language Processing.
Created by: Marvin Luckianto
This research paper introduces a novel approach for identifying toxic language in audio-based Indonesian content using Recurrent Neural Network (RNN), Bidirectional Long Short-Term Memory (BiLSTM), speech recognition, and natural language processing techniques. The proposed methodology transcribes Indonesian audio into text and employs natural language processing methods to extract lexical, syntactic, and semantic features to identify and categorize toxic language. Achieving high accuracy in detecting toxic language in Indonesian audio recordings, this approach outperforms existing methods. The RNN and BiLSTM model architecture captures the temporal dependencies of verbal content in audio recordings and gathers relevant information for toxicity classification. The paper reports a 95.2% accuracy, 96.4% precision, and 93.2% recall in identifying toxic speech recordings in the Indonesian language. The speech recognition component plays a crucial role in transcribing and classifying content. This technique can be applied in real-world scenarios such as content moderation in Indonesian social media platforms and detecting toxic language in customer service interactions, addressing the growing issue of toxic language in Indonesian online communities and social media platforms.