基于改进型支持向量机的语音信号情感识别研究

Research on Speech Signal Emotion Recognition Based on Improved Support Vector Machine

语音信号包含了人类丰富的情感信息,本文从离散情感模型出发,选择了高兴、悲伤、愤怒和害怕4种基本情感作为研究对象,利用萤火虫算法实现了支持向量机参数自动寻优,从而提高了识别的效率。为了使测试数据更据代表性,选取了中文和德文数两种据库,400个样本进行训练和测试。实验表明,对SVM分类器进行优化在一定程度上提高了情感识别率。

Speech signal contains abundant human emotional information. Starting from the discrete emotional model, this paper chooses four basic emotions: happiness, sadness, anger and fear as the research object, and uses firefly algorithm to realize automatic parameter optimization of support vector machine, thus improving the efficiency of recognition. In order to make the test data more representative, 400 samples of Chinese and German databases were selected for training and testing. Experiments show that the optimization of SVM classifier improves the emotion recognition rate to a certain extent.