Multi-Level Attention-Based Categorical Emotion Recognition Using Modulation-Filtered Cochleagram

Speech emotion recognition is a critical component for achieving natural human–robot interaction. The modulation-filtered cochleagram feature based on auditory modulation perception, which contains multi-dimensional spectral–temporal representation. In this study, we propose an framework that utilizes multi-level attention network to extract high-level emotional representations from the cochleagram. Our approach channel-level and spatial-level modules generate saliency maps of channel spatial representations, capturing significant space 3D convolution maps, respectively. Furthermore, employ temporal-level module capture regions concatenated sequence maps. experiments Interactive Emotional Dyadic Motion Capture (IEMOCAP) dataset demonstrate significantly improves prediction performance categorical compared other evaluated features. Moreover, our achieves comparable unweighted accuracy 71% in by comparing with several existing approaches. summary, study demonstrates effectiveness speech recognition, proposed provides promising direction future research field.