Customer Voice Emotions Analysis Dataset (CVEAD)

The Customer Voice Emotions Analysis Dataset (CVEAD) is a novel speech emotion recognition dataset tailored for the customer service domain. Existing SER datasets used in this context are either proprietary or not publicly accessible, limiting reproducibility and progress in real-world applications. Additionally, most publicly available SER datasets are constrained to acted emotions or entertainment-based contexts. CVEAD addresses these gaps by combining two data sources: (i) mock customer service calls extracted from YouTube, and (ii) scripted emotional utterances recorded by 12 volunteers based on realistic customer service scenarios. The dataset includes four emotion categories, Angry, Happy, Sad, and Neutral. All audio clips were annotated through majority voting by five independent annotators. Transcriptions were generated using OpenAI\u2019s Whisper model and manually corrected by two reviewers to ensure accuracy. As a publicly available resource, CVEAD supports the development and evaluation of SER systems in customer service environments.