AVEC 2011–The First International Audio/Visual Emotion Challenge

Abstract

Abstract. The Audio/Visual Emotion Challenge and Workshop (AVEC 2011) is the first competition event aimed at comparison of multimedia processing and machine learning methods for automatic audio, visual and audiovisual emotion analysis, with all participants competing under strictly the same conditions. This paper first describes the challenge participation conditions. Next follows the data used -the SEMAINE corpus -and its partitioning into train, development, and test partitions for the challenge with labelling in four dimensions, namely activity, expectation, power, and valence. Further, audio and video baseline features are introduced as well as baseline results that use these features for the three sub-challenges of audio, video, and audiovisual emotion recognition.

Modeling Latent Discriminative Dynamic of Multi-dimensional Affective Signals

“…The problem with this approach is that the feature space will end up extremely large (5900 dimensions of visual and 1941 of audio features in the case of Schuller et al [21]). This high dimensionality issue can be partially solved by performing dimensionality reduction or feature selection.…”

“…For all our experiments we used the dataset provided by Schuller et al [21]. The dataset consist of 95 video and audio recorded dyadic interaction sessions between human participants and a virtual agent operated by a human.…”

“…Our model is evaluated on the First International Audio/Visual Emotion Challenge (AVEC 2011) dataset. A complete description of the challenge and the dataset can be found in Shuller et al [21]. For the challenge the originally continuous dimensions were redefined as binary ones based on whether they were above or below average, this reduced a regression problem into a classification one.…”

Naturalistic Affective Expression Classification by a Multi-stage Approach Based on Hidden Markov Models

“…For the audio dataset, each uttered word is described by a vector of 1941 features and a set of four labels representing the level of activation, valence, expectation and power. The detailed description of the features can be found in [13]. In order to reduce the dimension of the feature vector, PCA was used and only 100 principle components were selected in the following experiments as they covered most of the variance.…”

“…Naturalistic expressions, differently from acted ones, change slowly as a person interacts with the environment. The AVEC challenge [13] provides a unique dataset of naturalistic audio and facial expressions to help address this issue. These data have been recorded at a high sampling rate making it possible to capture and analyze the slow transition between affective expressions.…”