基于时频上下文信息提取的心理诊断机器人情感识别研究

Research on Emotion Recognition of Psychological Diagnosis Robot Based on Time Frequency Context Information Extraction

为提高心理诊断机器人情感识别的准确率,提出一种基于时频上下文信息的语音情感识别方法。首先,采用卷积神经网络CNN对语音信息的时域上下文特征进行提取;然后引入长短时记忆网络LSTM识别语音频域的上下文情感信息特征;最后提出融合时频上下文信息方法,将提取的时间域和频率域上下文信息特征进行融合,以此应用到心理诊断机器人中进行情感识别。实验结果表明,相较于基于时域的上下文特征CFTD和基于频域的上下文特征CFFD方法的语音情感识别结果,提出的融合语音情感识别方法的情感识别平均正确率保持在96%左右,最高可达99.7%,相较于前两种识别方法的平均识别率分别高出了13.9%和16.6%。由此说明,提出的方法可有效提升语音情感的时域、频域上下文特征识别率,将此方法应用到心理诊断机器人的后端模块后,可实现机器人的语音情感准确识别,增强了心理诊断机器人的诊断效果。

To improve the accuracy of emotion recognition in psychological diagnostic robots,a speech emotion recognition method based on time-frequency contextual information is proposed.Firstly,the convolutional neural network CNN is used to extract the temporal context features of speech information;Then,the long and short term memory network LSTM is introduced to recognize contextual emotional information features in the speech frequency domain;Finally,a fusion method of time-frequency contextual information is proposed,which integrates the extracted temporal and frequency domain contextual information features and applies them to emotional recognition in psychological diagnostic robots.The experimental results show that compared to the speech emotion recognition results of the time-domain based context feature CFTD and frequency-domain based context feature CFFD methods,the proposed fusion speech emotion recognition method maintains an average emotion recognition accuracy of about 96%,with a maximum of 99.7%,which is 13.9%and 16.6%higher than the average recognition rates of the first two recognition methods,respectively.This indicates that the proposed method can effectively improve the recognition rate of temporal and frequency-domain contextual features of speech emotions.By applying this method to the backend module of the psychological diagnosis robot,accurate recognition of the robot's speech emotions can be achieved,enhancing the diagnostic effectiveness of the psychological diagnosis robot.