摘要
创建一个中文语音识别模型,采用近似深度全序列卷积神经网络(DFCNN)、门控循环单元(GRU)和连接时序分类(CTC)结构,将语音信号转换为二维语谱图像信号,将语音转换成中文文本。并在此模型的基础上,通过Keras编程搭建HTTP协议语音识别服务器,提供语音识别API,实现语音识别功能。测试结果表明,该方法能获得平均80%以上的语音识别准确率,取得了较好应用效果。
A Chinese speech recognition model is created.The structure of approximate deep fully convolutional neural network(DFCNN),gated recurrent unit(GRU)and connectionist temporal classification(CTC)are adopted.The speech signal is converted into two-dimensional speech spectra,converting speech into Chinese text.Based on this model,the speech recognition server of HTTP protocol is built by keras programming.Speech recognition API is provided to realize speech recognition function.The test results show that the method can achieve over 80%in average speech recognition accuracy.Good application results have been obtained.
作者
杨焕峥
YANG Huan-zheng(Jiangsu Research and Development Center of Application Technology for Wireless Sensing System,Wuxi,Jiangxi,China 214153;Wuxi Institute of Commerce,Wuxi,Jiangsu,China 214153)
出处
《湖南邮电职业技术学院学报》
2020年第3期24-27,共4页
Journal of Hunan Post and Telecommunication College
基金
2020年江苏高校“青蓝工程”资助项目“优秀青年骨干教师培养对象”
2018年无锡商业职业技术学院卓越师资队伍建设项目“校级教学名师培育对象”(项目编号:RS18MP03)
2018年无锡商业职业技术学院卓越师资队伍建设项目“校级骨干教师培育对象”(项目编号:RS18GG38)。
关键词
深度学习
语音识别
应用程序接口
Keras编程
deep learning
speech recognition
application program interface
Keras programming
