End-to-end speech emotion recognition using 3-d convolutional recurrent neural networks based on modulation spectral features

摘要

In this paper, we studied auditory-inspired end-toendrnrecognition of emotional speech using a 3-drnCRNN model based on ST modulation. Thernexperimental results demonstrate that our method isrnan effective way to design an emotion recognitionrnsystem by mimicking the human auditory system. Inrnthe future, we will compare Gammachirp and thernGammatone filter with a different number of acousticrnand modulation channels.