72‐3: Fully Convolutional Transformer‐Based Speech Emotion Recognition for Automotive Systems

We introduce a fully convolutional transformer for speech emotion recognition with application to automotive systems. The proposed architecture is composed of convolutional channel expansion, multi‐head attention and feed‐forward layers. We employ a trainable emotion query to better capture the characteristics of different emotions. In addition, we consider channel attention to better enable real‐time processing. Experiments show that the proposed method provides better performance than the benchmark algorithms.