IndEmoVis: An Indian Multimodal Audio-Visual Emotion Dataset for Conversational Interactions

The dataset comprises a total of 122 videos, involving 61 participants, consisting of 25 females and 36 males, with ages ranging from 18 to 21 years. Each video captures a conversational interaction between two participants, with two cameras mounted on the table to record separate videos of the speaker and the listener. These recorded videos are subsequently segmented into clips, each showcasing different emotions expressed by the participants. The dataset comprises a total of 122 videos, involving 61 participants, consisting of 25 females and 36 males, with ages ranging from 18 to 21 years. Each video captures a conversational interaction between two participants, with two cameras mounted on the table to record separate videos of the speaker and the listener. These recorded videos are subsequently segmented into clips, each showcasing different emotions expressed by the participants. The data collection process involved conversational interactions between pairs of participants within a controlled environment. The recordings captured a combination of spontaneous and guided conversations. The dataset consists of 45 recordings of guided conversations and 77 recordings of natural conversations, resulting in a total of 216 video clips portraying various emotions from the guided interactions and 379 video clips displaying emotions from the natural conversations. To facilitate further analysis and research, the video clips were further processed to extract image frames. From each clip, the frame with the highest intensity, capturing the peak expression of the corresponding emotion, was selected for subsequent analysis. These peak images serve as representative snapshots of the emotional expressions and can be used for facial emotion recognition tasks. IndEmoVis dataset is annotated for  six basic emotion categories (Happiness, Sadness, Surprise, Disgust, Anger, Fear), along with complex emotion categories of Awe and Sympathy, and a Neutral emotion class. Additionally, the dataset is annotated for the presence of braces, spectacles, and prominent hand gestures, providing valuable context for emotion analysis. Furthermore, intensity and confidence levels are annotated on a scale of 0 to 5 and 0 to 7, respectively, by the clinical psychologist to further enrich the dataset with nuanced emotional expressions.