audio-visual data set containing around 81 hours of conversational data among groups of 8-12 people playing a role game.

Get Data

Practical description of the data

The ''Werewolf'' data set consists of audio-visual recordings of 15 games played by 4 groups of people. The participants were volunteers from the research community at the Idiap Research Institute. In total 36 different people were recorded for the data set. The four groups had sizes of 10, 8, 10 and 12 participants respectively.

Each group also had a discussion phase at the start of the session, so that players could familiarize themselves with the rules of the game. The video was recorded using three horizontally mounted cameras. Audio was recorded from head-mounted microphones and an array microphone placed at the center at 48kHz. The maximum duration of a day-phase was set to be 2 x (No. of players).

This gave us 81.17 hours of audio-visual data. In total, the recordings lasted for 7.3 hours.

The figure below provides an example of the room layout for one of the groups. A more detailed description of the data can be found in the technical report that accompanies this database.


Sharing Annotations

This is a description of how to share your data.

As with any data set, the more we share, the more we can make the state-of-the-art better. The data has potential to be annotated for many things. This data set is made freely available by the generous volunteers who gave up their free time to help out with the recordings. To make their efforts worth while, please publish your annotations once you have published experimental results with the data. You can provide annotations for this data set by contacting the data manager (data-manager at idiap.ch).


If you use this data, please cite the following paper:

H. Hung and G. Chittaranjan, "The Wolf Corpus: Exploring group behaviour in a competitive role-playing game", ACM Multimedia, Florence, Italy, 2010.