Audio-based Indonesia Toxic Language Classification using Bidirectional Long Short Term Memory

Users are able to connect with one another via web-based internet technologies through social media platforms like Twitter. On these platforms, hate speech has become a troubling problem, when people or groups are targeted because of characteristics like color, ethnicity, gender, nationality, religion, or organization. Such information is spreading quickly as a result of the practice of using hate speech to gain fast notoriety. 80% of cybercrime prosecutions in Indonesia involve hate speech and defamation on social media, despite the fact that there are regulations like the UU ITE Law, Chapter 28, Number 2, that forbid inciting hatred or enmity on the basis of race, religion, or ethnicity. This research paper explores a novel approach to audio-based toxic language classification in the Indonesian language, leveraging key techniques including Bidirectional Long Short-Term Memory (BiLSTM), speech recognition, and natural language processing. The study focuses on identifying toxic language using a balanced dataset of 5000 entries, equally divided between toxic and non-toxic categories. This dataset was compiled from diverse sources such as NusaCrowd, Kaggle, and others, and was subsequently cleansed of emojis and symbols to maintain a plain text environment. Utilizing audio-based content analysis, the methodology involved the transcription and classification of audio data, yielding an accuracy of 95.2%, precision of 96.4%, recall of 93.2% and F1 score of 94.8% in the detection of toxic speech. The implications of this research extend beyond the academic sphere of toxic language detection, holding significant real-world application potential in areas such as content moderation for Indonesian social media platforms and toxic language detection in customer service interactions. Thus, this study addresses a critical issue of toxic language proliferation in Indonesian online communities and social media platforms, further contributing to the development of healthier digital environments in the future.