On-line Emotion Recognition in a 3-D Activation-Valence-Time Continuum using Acoustic and Linguistic Cues

摘要: For many applications of emotion recognition, such as virtual agents, the system must select responses while user is speaking. This requires reliable on-line recognition user’s affect. However most systems are based on turnwise processing. We present a novel approach to from speech using Long Short-Term Memory Recurrent Neural Networks. Emotion recognised frame-wise in two-dimensional valence-activation continuum. In contrast current state-of-the-art approaches, performed low-level signal frames, similar those used for recognition. No statistical functionals applied feature contours. Framing at higher level therefore unnecessary and regression outputs can be produced real-time every input frame. also investigate benefits including linguistic features frame obtained by keyword spotter.