Towards Affect Recognition: an Ica Approach

The speech signal can be thought of as multi-spatial signal where each signal subspace corresponds to different attributes that affect the speech. Language, accent, speaker, emotions are few potential sub-spaces. Each space can further be projected into n-dimensions, as there are different languages, accents and emotions, which combine together to make their sub-space. In this paper, we investigate detecting emotions in speech by projecting an emotional sub-space into a 2-dimensional space. The basis of this space is neutral and stress emotions. A previously formulated stress dependent TEO based feature is employed. To identify the weights of each component we use independent component analysis. This approach was used to exploit the non-Gaussianity of the data. Evaluations are conducted on the SOQ stress database using ICA sub-space decomposition. The results here suggest that dimensionality reduction offers a promising new approach to TEO based stress classification.