MuSe-Humor: Cross-Cultural Humor Detection Sub-Challenge (MuSe 2023)

<strong>Description</strong>: Predicting the presence of humor in football press conference recordings. Available modalities: audio and video, transcriptions. The dataset offers a cross-cultural setting: all press conferences in the training and development partition are held in German, all videos in the test partition are taken from English Premier League press conferences. <strong>Important</strong>: <strong>this package does not include the recordings themselves, please find them here</strong>: https://zenodo.org/record/7843460 <strong>Labels</strong>: windows of 2 seconds are given a binary label, indicating presence and absence of humor. At least 7 human annotators labelled each video. <strong>Dataset</strong>: MuSe-Humor is based on the Passau-Spontaneous Football Coach Humor (Passau-SFCH) dataset (paper). It includes press conference recordings of 10 different German Bundesliga football coaches collected in 2017. Moreover, for the MuSe 2023 challenge, recordings of 6 Premier League coaches have been added. The data provided here only includes segments in which the respective coach is speaking. The training data set contains recordings of 7 German-speaking coaches, development data consists of videos of 3 German-speaking coaches; and the test data set comprises recordings of 6 English-speaking coaches (one native speaker). <strong>General</strong>: The 4th Multimodal Sentiment Analysis Challenge and Workshop (MuSe) 2023 adresses research questions that are of interest to affective computing, machine learning and multimodal signal processing communities and encourages a fusion of their disciplines. The goal of the MuSe workshop and challenge is to gain new insights into the merits of each of the core modalities and to serve as a stimulating environment for the development and evaluation of multimodal affect recognition approaches.**描述**：预测足球新闻发布会录音中的幽默表达。可用模态涵盖音频、视频及转写文本。该数据集采用跨文化设置：训练集与开发集的所有发布会均使用德语，而测试集的所有视频均取自英格兰足球超级联赛（英超）的新闻发布会。 **重要提示**：本数据包未包含原始录音文件，原始录音可通过以下链接获取：https://zenodo.org/record/7843460 **标注规则**：将数据划分为2秒的时间窗口，并为每个窗口赋予二分类标签，用以标记该窗口内是否存在幽默表达。每条视频均至少由7名人工标注员完成标注。 **数据集概况**：MuSe-Humor数据集基于帕绍自发足球教练幽默（Passau-Spontaneous Football Coach Humor）数据集（详见相关学术论文）构建。其初始数据采集自2017年的10位德国足球甲级联赛（德甲）主教练的新闻发布会录音。为适配2023年多模态情感分析挑战赛与研讨会（Multimodal Sentiment Analysis Challenge and Workshop，简称MuSe），主办方新增了6位英超主教练的发布会录音。本数据集仅包含对应教练的发言片段。其中，训练集包含7位德语籍主教练的录音数据，开发集包含3位德语籍主教练的视频数据；测试集则涵盖6位英语籍主教练的录音数据（其中1位为英语母语使用者）。 **概述**：第四届多模态情感分析挑战赛与研讨会（MuSe 2023）聚焦情感计算、机器学习及多模态信号处理领域的研究议题，旨在推动各学科的交叉融合。本次研讨会与挑战赛的核心目标在于深入探究各核心模态的应用价值，并为多模态情感识别方法的研发与评估提供富有启发性的交流与验证平台。