摘要
智能语音技术包含语音识别、自然语言处理、语音合成三个方面的内容,其中语音识别是实现人机交互的关键技术,识别系统通常需要建立声学模型和语言模型。神经网络的兴起使声学模型数量急剧增加,基于神经网络的声学模型与传统识别模型相结合的方式,极大地推动了语音识别的发展。语音识别作为人机交互的前端,具有许多研究方向,文中着重对语音识别任务中的文本识别、说话人识别、情绪识别三个方向的声学模型研究现状进行归纳总结,尽可能对语音识别技术的演化进行细致介绍,为以后的相关研究提供有价值的参考。同时对目前语音识别的主流方法进行概括比较,介绍了端到端的语音识别模型的优势,并对发展趋势进行分析展望,最后提出当前语音识别任务中面临的挑战。
Intelligent speech technology includes speech recognition,natural language processing and speech synthesis.Speech recognition is a key technology for human-computer interaction,and the acoustic model and language model are usually needed to establish for recognition system.The rise of neural network leads to a sharp increase in acoustic models.The combination of acoustic models based on neural network and traditional recognition models greatly promotes the development of speech recognition.As the front end of human-computer interaction,speech recognition has many research directions.In this study,we mainly summarize the current research status of acoustic models in three directions of text recognition,speaker recognition and emotion recognition,and make a detailed introduction of the evolution of speech recognition technology as far as possible,so as to provide valuable reference for the related research in the future.At the same time,we generalize and compare the main methods of speech recognition,introduce the advantages of the end-to-end speech recognition model,analyze the development trend and present the challenges in the current speech recognition tasks at the end.
作者
叶硕
褚钰
王祎
李田港
YE Shuo;CHU Yu;WANG Yi;LI Tian-gang(Wuhan Research Institute of Posts and Telecommunications,Wuhan 430000,China)
出处
《计算机技术与发展》
2020年第3期181-186,共6页
Computer Technology and Development
基金
2018年度湖北省科学技术创新专项重大项目(2018AAA063)。
关键词
语音识别
声学模型
神经网络
深度学习
speech recognition
acoustic model
neural network
deep learning
