摘要
情感识别是人机交互领域中必须解决的关键问题。针对语音情感的识别问题,文中把遗传算法和小波神经网络算法相结合,即利用遗传算法具有的高度并行、随机、自适应搜索性能来选取初值进行训练,用小波神经网络来完成给定精度的学习。这样在解决复杂和非线性问题时,具有明显的优势。文中主要研究了四种基本的人类情感:喜悦、愤怒、悲伤和恐惧。并与BP算法和小波神经网络算法进行了比较,实验结果表明,该模型不但能够提高情感识别的正确率,缩短系统识别时间,而且为算法的实用性奠定了基础。
Emotion recognition is the key issue that must be solved in human-computer interaction. Aiming at the recognition problem of speech emotion, combined genetic algorithm and wavelet.neural network algorithm, utilize the performance of height parallel, random and adaptive search to select initial values, and using wavelet neural network to finish the learning. This has obvious advantages of solving complex and nonlinear problem. Four basic human emotions including joy, anger, sadness and fear were studied, and compared with back propagation neural network （BPNN） and wavelet neural network. The experimental results indicate that this method effectively improves the correct rate of emotion recognition, shortens the system recognition time, and lays the foundation for algorithm practicality.
出处
《计算机技术与发展》
2013年第1期75-78,共4页
Computer Technology and Development
基金
国家自然科学基金资助项目(60974071)
辽宁省自然科学基金(201102005)
关键词
情感识别
神经网络
遗传算法
小波分析
emotion recognition
neural network
genetic algorithm
wavelet transform
