A Multimodal Depression Consultation Dataset of Speech and Text with HAMD-17 Assessments

Abstract

The global surge in depression rates, notably severe in China with over 95 million affected, underscores a dire public health issue. This is exacerbated by a critical shortfall in mental health professionals, highlighting an urgent call for innovative approaches. The advancement of Artificial Intelligence (AI), particularly Large Language Models, offers a promising solution by improving mental health diagnostics. However, there is a lack of real data for reliable training and accurate evaluation of AI models. To this end, this paper presents a high-quality multimodal depression consultation dataset, namely Parallel Data of Depression Consultation and Hamilton Depression Rating Scale (PDCH). The dataset is constructed based on clinical consultations from Beijing Anding Hospital, which provides audio recording and transcribed text, as well as corresponding HAMD-17 scales annotated by professionals. The dataset contains 100 consultations and the audio exceeds 2,937 minutes. Each of them is about 30-min long with more than 150 dialogue turns. It enables to fill the gap in mental health services and benefit the creation of more accurate AI models.