基于多模态的呼叫中心质量监控系统的研究和实现

发布时间：2018-06-24 13:01

本文选题：呼叫中心 + 关键词检出　；参考：《浙江工商大学》2015年硕士论文

【摘要】：近几年,呼叫中心迅速发展,在企业中所占的地位越来越重要。呼叫中心是企业和客户之间的重要沟通渠道。呼叫中心的语音质量就直接影响了企业和客户关系。随着经济的发展,消费理念的变化,客户更加关心自身的权益,当自身的权益受到侵犯或者对自身的权益不了解的情况下,客户就会通过呼叫中心了解自身权益。如果这时话务员的服务质量太差将直接影响了客户对企业的印象。随着呼叫中心的发展,客户消费理念的变化,呼叫中心的话音数量将会急剧增长。面对话音数量的增长,传统的人工模式已经不能满足现在的状况。基于上述情况,利用关键词检出技术和情感分析技术实现话音质量的自动化监控,以提高服务质量缓解企业和客户的关系。本文首先介绍了话务服务质量监控系统的两个模块：语音关键词检出和语音情感分析。关键词检出模块主要包括语音预处理、特征提取、声学模型训练、关键词检出。声学模型的训练是在现有声学模型的基础上进行自适应训练,以使声学模型在呼叫中心领域内有更好的识别率。使用SphinxTrain工具进行声学模型的自适应训练。关键词检出主要是使用由CMU开源的PocketSphinx工具实现,采用的是填充模型。语音情感识别模块采用人工神经网络作为情感分类器。本文主要介绍了模型训练算法和情感分类算法。重点介绍了反向传播算法用于神经网络训练。本文在关键词检出和语音情感识别的基础上设计了话音服务质量评测指标,根据指标设计了话音服务质量评测算法并将其集成到系统中。通过查看话音服务质量评测结果,呼叫中心的经理和话务员都可以快速定位优缺点,提高话音服务质量。
[Abstract]:In recent years, call centers have developed rapidly and become more and more important in enterprises. Call center is an important communication channel between enterprises and customers. The voice quality of call center directly affects the relationship between enterprise and customer. With the development of economy and the change of consumption concept, customers are more concerned about their own rights and interests. When their rights and interests are violated or their rights and interests are not understood, customers will know their rights and interests through the call center. If the operator's service quality is too poor will directly affect the customer impression of the enterprise. With the development of call center and the change of customer consumption concept, the number of voice in call center will increase rapidly. In the face of the increase in the number of voice, the traditional manual mode can not meet the current situation. Based on the above situation, the automatic monitoring of voice quality is realized by using keyword detection technology and emotion analysis technology, in order to improve the quality of service and alleviate the relationship between enterprises and customers. This paper first introduces two modules of the quality of service monitoring system: voice keyword detection and voice emotion analysis. Keyword detection module mainly includes speech preprocessing, feature extraction, acoustic model training, keyword detection. The training of acoustic model is based on the existing acoustic model, so that the acoustic model has a better recognition rate in the field of call center. The Sphinx Train tool is used for adaptive training of acoustic models. Keyword detection is mainly implemented by the open source Pocket Sphinx tool of CMU, and the filling model is used. The speech emotion recognition module uses artificial neural network as emotion classifier. This paper mainly introduces the model training algorithm and emotion classification algorithm. The application of back propagation algorithm to neural network training is emphasized. Based on keyword detection and speech emotion recognition, this paper designs voice quality of service evaluation index, designs voice quality of service evaluation algorithm according to the index and integrates it into the system. By checking the results of voice service quality evaluation, call center managers and operators can quickly locate their advantages and disadvantages and improve the quality of voice service.
【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP277

【相似文献】