Data

Hume-Mimic

In the MuSe-Mimic sub-challenge, the the novel Hume-Mimic database is utilized. Hume-Mimic is a large-scale multimodal database containing over 20 hours of user-generated content by more than 500 subjects. Each of the over 18,000 video clips in MuSe-Mimic shows a user mimicking an emotion. The videos are labelled for three emotions: Approval, Disappointment and Uncertainty.

The data is split into training, development and test partition in a speaker independent way. 

Passau-SFCH

The MuSe-Humor sub-challenge uses the novel Passau Spontaneous Football Coach Humor (Passau-SFCH) database. It comprises audiovisual recordings of German and English football Bundesliga press conferences. It is annotated for humor displayed by the coaches. For the challenge, a binary labelling (presence or absence of humor) is provided. 

Overall, 10 hours of recordings from 10 different coaches are contained in the training and development set. The test partition comprises about 6.5 hours of English Premier League press conferences held by 6 different coaches. [paper] 

Ulm-TSST

For MuSe-Personalisation, the Ulm-TSST database is used, supplying a multimodal dataset annotated with valence and arousal signals. It features recordings of individuals undergoing the stress inducing TSST scenario. In addition to the audiovisual recordings, Ulm-TSST includes biological recordings, such as Electrocardiogram (ECG),  Electrodermal Activity (EDA), Respiration, and Heart Rate (BPM). With 69 participants (69.5% female) aged between 18 and 39 years, a total of about 6 hours were accumulated. For MuSe-Personalisation, parts of the test labels are released in addition to the training and development data, thus facilitating the development of methods tailored to specific individuals.