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

摘要

In naturalistic behaviour, the affective states of a person\udchange at a rate much slower than the typical rate at which video or\udaudio is recorded (e.g. 25fps for video). Hence, there is a high probability\udthat consecutive recorded instants of expressions represent a same\udaffective content. In this paper, a multi-stage automatic affective expression\udrecognition system is proposed which uses Hidden Markov Models\ud(HMMs) to take into account this temporal relationship and finalize the\udclassification process. The hidden states of the HMMs are associated\udwith the levels of affective dimensions to convert the classification problem\udinto a best path finding problem in HMM. The system was tested on\udthe audio data of the Audio/Visual Emotion Challenge (AVEC) datasets\udshowing performance significantly above that of a one-stage classification\udsystem that does not take into account the temporal relationship, as well\udas above the baseline set provided by this Challenge. Due to the generality\udof the approach, this system could be applied to other types of\udaffective modalities.