Speech Emotion Recognition Based on Deep Residual Shrinkage Network

Speech emotion recognition (SER) technology is significant for human–computer interaction, and this paper studies the features modeling of SER. Mel-spectrogram introduced utilized as feature speech, theory extraction process mel-spectrogram are presented in detail. A deep residual shrinkage network with bi-directional gated recurrent unit (DRSN-BiGRU) proposed paper, which composed convolution network, unit, fully-connected network. Through self-attention mechanism, DRSN-BiGRU can automatically ignore noisy information improve ability to learn effective features. Network optimization, verification experiment carried out three emotional datasets (CASIA, IEMOCAP, MELD), accuracy 86.03%, 86.07%, 70.57%, respectively. The results also analyzed compared DCNN-LSTM, CNN-BiLSTM, DRN-BiGRU, verified superior performance DRSN-BiGRU.