Prototype Development of Face and Speaker Recognitions based on Edge Computing
Yuxuan Pan, Xiaowen Peng, Xuexian Lin, Minghua Xia · 2020
In recent years, cognitive services like face recognition and speech recognition has found wide applications in smart cities. The main approach of providing cognitive services depends on cloud computing, which however suffers high response latency and information security problems. In this paper, based on edge computing, prototypes on face recognition and speaker recognition are developed on a cloud radio access network platform. In the face recognition architecture, a traditional face recognition network is cascaded to a face detection network so as to mitigate the interference of irrelevant information. In the speaker recognition architecture, multi-task learning is exploited to improve the generalization of neural network, and a phoneme recognition task is employed as an auxiliary task to assist the speaker recognition. Experimental results corroborate the effectiveness of the proposed architectures in terms of recognition accuracy and response latency.