MSP-Podcast

该数据集名为MSP-Podcast，包含了英语母语者说的短句，每个说话者的发音时长介于3至11秒之间。数据集中还包含了由多位评分者手动分配的情感值，包括愉悦度、激活度和支配度（采用7点李克特量表评分）。该数据集专门用于评估情感模型，并设有特定的分割（Eval1.3和Eval1.6），同时包含了在噪声和混响条件下的变体，以便进行鲁棒性分析。评分由多位评分者提供，任务是对话语音进行多维情感识别。

The dataset, named MSP-Podcast, comprises short utterances spoken by native English speakers, with each audio clip's duration ranging from 3 to 11 seconds. It also includes emotion-related scores manually assigned by multiple raters, covering valence, arousal, and dominance, which are rated on a 7-point Likert scale. This dataset is specifically developed for evaluating emotion models, and it features predefined splits (Eval1.3 and Eval1.6). Additionally, it contains variants under noisy and reverberant conditions to facilitate robustness analysis. The scores are provided by multiple raters for the task of multi-dimensional emotion recognition on conversational speech.