摘要
为了改善呼叫中心的客户体验和服务质量,首先构建了一个面向呼叫中心的语音情感识别数据集。数据集采集自呼叫中心真实的业务数据,总计时长5小时16分钟,其中包含客服样本3008条,用户样本1486条,总计4494条语音样本。与其它语音情感识别数据集相比,该数据集不仅包含了正向情绪、正常情绪和负向情绪三类离散情感标签,还包含了愉悦度和激活度两个维度的情感信息。然后,使用了基于CNN与LSTM-Transformer的语音情感识别模型,并采用焦点损失函数进行监督训练。实验表明,焦点损失函数可有效解决类别不平衡带来的偏差问题,使上述模型在本数据集上的识别正确率达到了94%。以上研究不仅丰富了中文语音情感识别的语料库,也为呼叫中心服务质检提供了更可靠的基础。
To improve the customer experience and service quality of call centers,this study first constructs a voice emotion recognition dataset for call centers.The dataset is collected from real business data of call centers,with a total duration of 5 hours and 16 minutes.It includes 3008 samples from customer service representatives and 1486 samples from users,making a total of 4494 voice samples.Compared to other voice emotion recognition datasets,this dataset not only contains three discrete emotion labels(positive emotion,neutral emotion,and negative emotion),but also includes two dimensions of emotion information:pleasantness and activation.Then,this study adopts a CNN with LSTM-Transformer-based voice emotion recognition model and uses the focal loss function for supervised training.The experimental results show that the focal loss function can effectively address the bias issue caused by class imbalance,achieving a recognition accuracy of 94%on this dataset.This research enriches the corpus of Chinese voice emotion recognition and provides a more reliable foundation for quality inspection in call center services.
作者
何坤成
王以松
李建州
张辉
HE Kun-cheng;WANG Yi-song;LI Jian-zhou;ZHANG Hui(College of Computer Science and Technology,Guizhou University,Guiyang Guizhou 550025,China;Postdoctoral Scientific Research Station,ShijiHengtong Technology Co.,Ltd,Guiyang Guizhou 550014,China;School of Information,Guizhou University of Finance and Economics,Guiyang Guizhou 550025,China;Intelligent Perception and Fusion Computing Engineering Technology Research Center,Guizhou University of Finance and Economics,Guiyang Guizhou 550025,China)
出处
《计算机仿真》
2025年第6期290-296,共7页
Computer Simulation
基金
贵州省高等学校智能计算与监测预警技术创新团队(黔教技[2023]063)
贵州省科技计划项目(黔科合支撑[2023]一般372)。
关键词
呼叫中心
语音情感识别
数据集
维度情感
融合模型
Call center
Voice emotion recognition
Dataset
Dimensional emotions
Fusion model
