Subtitle-Aligned Movie Sounds (SAM-S)

Subtitle-Aligned Movie Sounds (SAM-S)数据集由南加州大学信号分析与解释实验室创建，包含从430部电影中自动提取的超过11万条音频事件数据。该数据集利用公开可用的闭路字幕转录，通过自动化方法挖掘音频事件，并根据声音、来源和质量三个维度进行分类，最终形成包含245种声音的分类体系。创建过程中，研究团队采用了简单且可扩展的方法，确保了数据集的高精度和低人工干预。SAM-S数据集主要应用于音频事件检测领域，旨在提高音频识别的准确性和效率，特别是在电影等多媒体内容中的应用。

Subtitle-Aligned Movie Sounds (SAM-S) dataset was created by the Signal Analysis and Interpretation Laboratory of the University of Southern California. It contains over 110,000 audio event entries automatically extracted from 430 movies. Leveraging publicly available closed-caption transcripts, the dataset extracts audio events via automated methods, and classifies them along three dimensions: sound type, source, and quality, ultimately establishing a taxonomy consisting of 245 sound categories. During the dataset creation process, the research team adopted a simple yet scalable approach, ensuring high accuracy and minimal human intervention. The SAM-S dataset is primarily applied in the field of audio event detection, aiming to improve the accuracy and efficiency of audio recognition, especially for multimedia content such as movies.