论文部分内容阅读
为了检测工作人员的烦躁情绪,实现情感状态的评价,通过在工作环境中诱发情感语音,获取了足够的测试样本,建立了2 000条样本的工作环境情感语音数据库.在检测烦躁情绪过程中,首先提取语音的韵律特征和音质特征参数,然后利用基于蛙跳算法的改进的BP神经网络进行烦躁情绪识别.实验比较了BP,RBF和SFLA神经网络的性能,结果显示SFLA神经网络的识别率比BP神经网络高4.7%,比RBF神经网络高4.3%.实验结果表明,使用蛙跳算法训练随机初始数据可以优化神经网络的连接权重和阈值,加快收敛速度,提高识别率.
In order to detect staff’s irritability and achieve emotional status evaluation, we obtained enough test samples by inducing emotional voice in the working environment and established a working environment emotional voice database of 2 000 samples.In the process of detecting irritability, Firstly, the prosodic features and the sound quality parameters of speech are extracted, and then an improved BP neural network based on leapfrog algorithm is used to identify irritability.Experiments are used to compare the performance of BP, RBF and SFLA neural networks. The results show that the recognition rate of SFLA neural network BP neural network is 4.7% higher and 4.3% higher than RBF neural network.Experimental results show that using frog-jump algorithm to train stochastic initial data can optimize the connection weight and threshold of neural network, accelerate the convergence speed and improve the recognition rate.