摘要
针对当前心理健康智能机器人对学前儿童的语音情感识别准确率低,导致无法有效地对儿童进行心理辅导监测的问题,提出构建一个基于注意力机制Attention+长短期神经网络LSTM的学前儿童情感识别模型。该模型采用LSTM神经网络对原始语音中的时序关系进行保留,并在LSTM基础上,加入注意力机制,将传统遗忘门、输出门转换为注意力门,得到基于深度注意力门的Attention+LSTM模型,通过此模型对儿童语音情感特征进行深度挖掘,从而实现儿童语音情感准确识别。通过实验分析发现,在CASIA、e NTERFACE和GEMEP三个儿童情感语料库中,本模型的平均识别率(UAR)分别为90.6%,81.5%和51%,比传统LSTM模型分别高出了0.6%、5.65%和2.5%。由此可知,相较于传统LSTM模型,本模型可提升学前儿童的语音情感识别率,此模型可在心理健康智能机器人中进行应用,可实现儿童心理有效辅导和监测。
In view of the current problem that the current mental health intelligent robot has a low accuracy of speech and emotion recognition for preschool children, which leads to the inability to effectively monitor psychological counseling for children, a preschool children emotion recognition model based on attention mechanism Attention + long and short-term neural network LSTM is proposed.The model adopts LSTM neural network to retain the timing relationship in the original speech, and on the basis of LSTM, add attention mechanism, the traditional forgotten door, output door into attention door, get Attention + LSTM model, through this model for children voice emotion features, so as to realize children’s voice emotion accurate recognition.Through experimental analysis, we found that the average recognition rate(UAR) of CASIA, e NTERFACE and GEMEP was 90.6%, 81.5% and 51%, respectively, which were 0.6%, 5.65% and 2.5% higher than the traditional LSTM model.It can be seen that compared with the traditional LSTM model, this model can improve the speech and emotion recognition rate of preschool children. This model can be applied in the mental health intelligent robot to realize the effective psychological counseling and monitoring of children.
作者
宋文婧
SONG Wenjing(Xi’an Traffic Enginering Institute,Xi’an 710000,China)
出处
《自动化与仪器仪表》
2022年第11期199-204,共6页
Automation & Instrumentation
基金
陕西省教育厅2021年度一般专项科研项目《乡村振兴背景下农村学前儿童问题行为研究》(21JK0223)。
关键词
学前儿童
心理健康监测
语音情感识别
注意力机制
LSTM
preschool children
mental health monitoring
speech and emotion recognition
attention mechanism
LSTM
