Self-attention transfer networks for speech emotion recognition

Background A crucial element of human-machine interaction, the automatic detection of emotional states from human speech has long been regarded as a challenging task for machine learning models. One vital challenge in speech emotion recognition(SER) is learning robust and discriminative representations from speech. Although machine learning methods have been widely applied in SER research, the inadequate amount of available annotated data has become a bottleneck impeding the extended application of such techniques(e. g., deep neural networks). To address this issue, we present a deep learning method that combines knowledge transfer and self-attention for SER tasks. Herein, we apply the log-Mel spectrogram with deltas and delta-deltas as inputs. Moreover, given that emotions are timedependent, we apply temporal convolutional neural networks to model the variations in emotions. We further introduce an attention transfer mechanism, which is based on a self-attention algorithm to learn long-term dependencies. The self-attention transfer network(SATN) in our proposed approach takes advantage of attention transfer to learn attention from speech recognition, followed by transferring this knowledge into SER. An evaluation built on Interactive Emotional Dyadic Motion Capture(IEMOCAP)dataset demonstrates the effectiveness of the proposed model.