面向语音与面部表情信号的情感可视化方法

Emotion visualization method for speech and facial expression signals

为了提高情感可视化的鲁棒性,该文提出了一种面向语音与面部表情信号的情感可视化方法。首先对获取的情感信号进行特征提取,并将其作为神经网络的输入,神经网络的输出即为相应的图案信息,然后通过图像生成模块生成可视化图像,实现对中性、高兴、愤怒、惊奇、悲伤和恐惧六种人类基本情感的可视化。该方法通过组合不同模式的情感特征进入一幅图像中,为人们创造了情感的可读模式,可以直观地展示情感的分类结果。仿真实验结果表明,仅通过语音信号进行可视化的平均正确率是78.0%,而通过该文方法的平均正确率是91.8%,具有良好的鲁棒性和易懂性。

In order to improve the robustness of emotion visualization, this paper proposes a new emotion visualization method for speech and facial expression signals. Firstly, extracts emotion feature parameters. Then makes the feature parameters as the input of neural network, the output of neural network is the corresponding pattern information, and then generates a visual image by image generation module, and finally accomplishes the visualization for six kinds of human emotion（neutral, joy,anger, surprise, sadness, fear）. This method creates emotion readable mode for people by combining the emotion features of different patterns into an image. That can visually show emotion classification results. The simulation results show that the average correct rate is 78.0% only through speech signal, while the average correct rate is 91.8% through the proposed method.That is robust and easy to understand.