基于神经网络和虚幻引擎的数字人客服系统

Digital Human Customer Service System Based on Neural Networks and Unreal Engine

为了改善传统智能客服在基于文本交互的情境下,用户体验较差的问题。基于人脸的三维重建赋予数字人真实面孔,同时结合虚幻引擎进行实时渲染,采用Rasa(用于构建对话机器人的开源机器学习框架)针对不同情景进行中文问答训练,经由TTS(Text To Speech)算法转化文本为音频,再通过唇形同步算法将音频映射为数字人面部变形权重,该权重最终用于驱动面部动画,实现了用户和高保真数字人语音对话的功能。经过在不同年龄段群体进行测试和统计,用户对系统的满意率高达96%,为提升智能客服的体验提供了优化方案。

In order to improve the problem of poor user experience in traditional intelligent customer service based on text interaction,this paper proposes to endow digital humans with real faces based on 3D reconstruction and use virtual engine for real-time rendering.Rasa(an open-source machine learning framework for building dialogue robots)is used to train Chinese Q&A for different scenarios.Text is transformed into audio by TTS(Text To Speech)algorithm,and then the audio is mapped to digital human facial deformation weights by lip synchronization algorithm.This weight is ultimately used to drive facial animation and realize the function of voice conversation between users and high-fidelity digital humans.After testing and statistics in different age groups,the user satisfaction rate is as high as 96%,providing an optimization solution for improving the experience of intelligent customer service.