MSP-IMPROV

该数据集名为MSP-IMPROV语料库，是一个多模态情感数据库，其设计理念是在控制词汇内容和情感的同时，力求录音的自然性。该数据库包含了根据不同情感诱发情境分为四类的语音表达：自然互动（NI）、他人即兴（OI）、目标即兴（TI）和目标朗读（TR）。此外，每个语音表达都根据情感诱发方法附有一个类别标签，并提取了包括梅尔滤波器组系数和能量在内的声学特征，以及利用预训练的面部识别模型得到的视觉特征嵌入。该数据集的任务是进行音频-视觉情感识别。

This dataset, named the MSP-IMPROV Corpus, is a multimodal emotion database designed to prioritize the naturalness of recordings while controlling lexical content and emotional attributes. It includes four categories of speech expressions categorized based on different emotion-eliciting scenarios: Natural Interaction (NI), Other-Initiated Improvisation (OI), Target-Initiated Improvisation (TI), and Targeted Reading (TR). Additionally, each speech expression is assigned a category label according to the emotion-elicitation method, with acoustic features including Mel-Frequency Cepstral Coefficients (MFCCs) and energy extracted, as well as visual feature embeddings obtained using a pre-trained facial recognition model. The task of this dataset is audio-visual emotion recognition.