Deep Learning Enabled High-Performance Speech Command Recognition on Graphene Flexible Microphones

• Published in: ACS applied electronic materials 2022-05, Vol.4 (5), p.2306-2312
• Main Authors: Zhang, Xin-Yu, Liu, Hang, Ma, Xiang-Yu, Wang, Zi-Cheng, Li, Guo-Peng, Han, Lei, Sun, Kuan, Yang, Qi-Sheng, Ji, Shou-Rui, Yu, Du-Li, Li, Yu-Tao, Ren, Tian-Ling
• Format: Article
• Language: English
• ISSN: 2637-6113; 2637-6113
• DOI: 10.1021/acsaelm.2c00125

Benefiting from outstanding electrical properties and excellent flexibility, graphene is widely used in novel sensors. However, the high price, complex manufacturing process, and especially the lack of algorithm analysis for graphene signals limit the development of graphene sensors. In order to overcome these problems, this paper demonstrates a high-performance solution for audio recognition based on a novel low-cost graphene flexible microphone. The graphene microphone is fabricated by laser-induced graphene which has a lower price and a simple process compared to traditional chemical methods. Not only the production of graphene microphones, this paper also innovatively uses deep learning to achieve speech command recognition on graphene microphones. The signal characteristics of the graphene flexible microphone are analyzed, and we propose a deep learning algorithm suitable for graphene microphones. One-dimensional convolutional neural network enables the high-performance audio recognition on graphene microphones which has huge advantages over traditional pattern recognition methods. The voices of 10 numbers and 20 sentences were collected 50 times to build a data set with a total of 1500 samples. For 10 numbers from 0 to 9, the deep learning model achieved an average correct rate of 98% which is far more than 84.5% of the traditional method. Finally, 20 sentences were used to test the performance of this solution under more vocabulary and the accuracy rate was 98.25%. With the help of deep learning, this solution already has the basic functions of a simple speech recognition system. This paper shows a complete application that brings graphene microphones from theoretical research to practical applications and provides reference for signal analysis and algorithm research of more novel sensors.