CNN-BiGRU Speech Emotion Recognition Based on Attention Mechanism

Abstract

The speech emotion recognition system can significantly improve the efficiency of human-computer interaction by accurately recognizing emotional information in speech. This system typically includes two main steps: speech feature extraction and emotion classification. In order to improve accuracy, this article uses MFCC features, short-term energy features, and short-term average zero crossing rate as model inputs, and introduces a convolutional neural network based on attention mechanism and a bidirectional gated loop unit (BiGRU). This method can effectively focus on useful information in speech features. Compared to the CNN-BiLSTM network based on attention mechanism and the CNN-GRU network based on attention mechanism, this method can effectively improve the accuracy of speech emotion recognition when conducting experiments on the Chinese sentiment corpus CASIA.