Graphics Reference
In-Depth Information
3.1 Data collection
We will put focus on two different data sets, namely the EmoRec II
collected at the Ulm University and the AVEC 2011 dataset.
3.1.1 EmoRec II
A simulation of a natural verbal human-computer interaction was
implemented as a Wizard-of-Oz (WOZ) experiment (Kelley, 1984).
The WOZ experiment allows the simulation of computer or system
properties in a manner such that subjects have the impression that they
are having a completely natural verbal interaction with a computer-
based memory trainer. The design of the memory trainer followed the
principle of the popular game “Concentration”. The variation of the
system behavior, in response to the subjects, was implemented via
natural spoken language, with parts of the subject's reactions taken
automatically into account.
In order to induce target emotions during the experiment, we
considered the following affective factors that are implemented as
natural language dialog:
￿ Delaying the response of a command
￿ Non-execution of the command
￿ Simulate incorrect speech recognition
￿ Offer of technical assistance
￿ Lack of technical assistance
￿ Propose to quit the game ahead of time
￿ Positive feedback
The procedure of emotion induction is structured in differentiated
experimental sequences (ESs) in which the user is passed through VAD
octants (valence: positive, negative, neutral; arousal: low, high, neutral;
dominance: low control, high control, neutral) by the investigator
(compare Figure 6). Audio, video and physiological data (namely
electromyography, skin conductance level) were recorded.
Within this study, we focus on the recognition of the emotional
octants in ES-4 and ES-6. The database comprises eight subjects with
an average age of 63.5 years.
3.1.2 AVEC 2011 Data Collection
The second data collection used in this study has been provided
within the Audio/Visual Emotion Challenge (AVEC) 2011 of the ACII
2011 workshop. Overall three sub-challenges were proposed: an audio
Search WWH ::




Custom Search