Multi-branch feature learning based speech emotion recognition using SCAR-NET

Speech emotion recognition (SER) is an active research area in affective computing. Recognizing emotions from speech signals helps to assess human behaviour, which has promising applications the of human-computer interaction. The performance deep learning-based SER methods relies heavily on feature learning. In this paper, we propose SCAR-NET, improved convolutional neural network, extract emotional features and implement classification. This work includes two main parts: First, spectral, temporal, spectral-temporal correlation through three parallel paths; then split-convolve-aggregate residual blocks are designed for multi-branch refined by global average pooling (GAP) pass a softmax classifier generate predictions different emotions. We also conduct series experiments evaluate robustness effectiveness SCAR-NET can achieve 96.45%, 83.13%, 89.93% accuracy datasets EMO-DB, SAVEE, RAVDESS. These results show outperformance SCAR-NET.