Speech Emotion Recognition Method Based on Improved Long Short-term Memory Networks

Donglin Jiang ； Hang Li ； Shoulin Yin

《International Journal of Electronics and Information Engineering》 12卷4期 (2020/12) Pp. 147-154

https://doi.org/10.6636/IJEIE.202012_12(4).01

摘要

The purpose of speech emotion recognition is to enable the computer to discover the current emotional state of human beings from human speech signals, so that the machine can understand the perceptual thinking of human beings, so that the computer has more humanized and complex functions. Speech emotion recognition is a typical pattern recognition problem, which usually includes three key steps: speech feature extraction, dimension reduction and classification. However, the existing methods lose a lot of original information. The recognition accuracy is not high. In this paper, we propose a speech emotion recognition method based on improved Long Short-term Memory Networks (LSTM). Features are extracted from the original signal by a convolutional neural network. And it creates a 2-layer Long Short-term Memory Networks. Compared to other speech emotion recognition methods, the proposed method obtains a better results.

關鍵字

CNN ； LSTM ； Speech Emotion Recognition