DeToxy

DeToxy是由印度理工学院马德拉斯分校电气工程系语音实验室和印度理工学院德里分校MIDAS实验室共同创建的大型多模态数据集，专注于英语口语中的毒性分类。该数据集包含超过200万条从公开可用的语音数据库中提取的语音和文本数据，旨在为交互式智能系统中的内容审核提供支持，特别是在在线音频/视频内容、游戏和客户服务等领域。DeToxy通过人工标注进行毒性检测，支持单模态或多模态模型的发展，以更好地捕捉语音的语义信息。数据集的创建涉及使用预训练的wav2vec-2.0模型获取缺失的文本转录，并通过两步方法进行标注，确保数据的质量和准确性。

DeToxy is a large-scale multimodal dataset jointly created by the Speech Lab of the Department of Electrical Engineering, Indian Institute of Technology Madras and the MIDAS Lab of the Indian Institute of Technology Delhi, focusing on toxicity classification in spoken English. This dataset contains over 2 million pairs of speech and text data extracted from publicly available speech corpora, aiming to support content moderation in interactive intelligent systems, particularly in domains such as online audio/video content, gaming, and customer service. DeToxy conducts toxicity detection via manual annotation, and supports the development of unimodal or multimodal models to better capture the semantic information of speech. The construction of this dataset involves utilizing the pre-trained wav2vec-2.0 model to acquire missing text transcriptions, and adopts a two-step annotation approach to ensure the quality and accuracy of the dataset.