基于移动设备的听障人特定语音识别训练系统
|
摘要
我国国内听障人士目前达到7 200万,他们普遍存在言语沟通障碍,大多只能依赖特殊教育学校的手语、助听器以及人工耳蜗实现和正常人沟通交流.结合听障者的自身特点设计了特定的语音识别系统,听障者通过点击界面的图标实现发音,用户根据声音信息实现模仿,系统根据检测到的数据进行评测,给出听障者的训练结果,实现了沟通和康复的有机结合.在此系统的基础上,为了提高语音输入的高效性,增加了基于语义的sem算法的文本预测,该算法模型对句子中出现的所有词进行了构建.在特定的语音识别的基础上完成了整体的构架设计,实现了训练模块和语音模块之间的交互.
At present,there are 72 million hearing-impaired people in China,and they generally have speech communication problems.They cannot communicate with normal people in speech communication,and most of them can only rely on special education schools and expensive medical expenses.For hearing-impaired people,auxiliary equipment based on mobile devices is even less.In this paper,a specific speech recognition system was designed in combination with the characteristics of the hearers.The hearers realize the pronunciation by clicking the icon on the interface,the users imitate according to the sound information,and the system evaluates the results according to the detected data,so as to achieve the organic combination of communication and rehabilitation.On the basis of this system,in order to improve the efficiency of speech input,the text prediction of sem algorithm based on semantics was added.Based on the specific speech recognition,the overall architecture design was completed and the interaction between the training module and the speech module was realized.
At present,there are 72 million hearing-impaired people in China,and they generally have speech communication problems.They cannot communicate with normal people in speech communication,and most of them can only rely on special education schools and expensive medical expenses.For hearing-impaired people,auxiliary equipment based on mobile devices is even less.In this paper,a specific speech recognition system was designed in combination with the characteristics of the hearers.The hearers realize the pronunciation by clicking the icon on the interface,the users imitate according to the sound information,and the system evaluates the results according to the detected data,so as to achieve the organic combination of communication and rehabilitation.On the basis of this system,in order to improve the efficiency of speech input,the text prediction of sem algorithm based on semantics was added.Based on the specific speech recognition,the overall architecture design was completed and the interaction between the training module and the speech module was realized.
引文
[1]马蓉蓉.辅助沟通系统(AAC)在孤独症儿童沟通障碍中的应用研究[D].杭州:浙江工业大学,2013.
[2]李建文,李岩.多通道皮肤听声器质量检测系统的研究[J].计算机测量与控制,2013,21(11):2941-2943.
[3]SHANNON C.Prediction and entropy of printed English[J].In Bell Technical Journal,1951,30:50-64.
[4]屈丹,张文林.基于本征音子说话人子空间的说话人自适应算法[J].电子与信息学报,2015,37(6):1350-1356.
[5]Gri FFITHS V,KALDI S,THOMPSON S.Mid-career teacher educators:comparing professional learning and academic content in England and Greece[J].Separation&Purification Technology,2010,39(3):181-188.
[6]POVEY D,GHOSHAL A,BOULIANNE G,et al.The kaldi speech recognition toolkit[C]//Proc of IEEE 2011 Workshop on Automatic Speech Recognition and Understanding,Waikoloa Hawaii,2011:128-131.
[7]谢怡宁,黄金杰,何勇军.噪声环境下智能机器人语音控制特征提取方法[J].北京邮电大学学报,2013,36(3):83-87.
[8]张建华.基于深度学习的语音识别应用研究[D].北京:北京邮电大学,2015.
[9]陆俊,张琼,杨俊安,等.嵌入深度信念网络的点过程模型用于关键词检出[J].信号处理,2013,7(12):865-872.
[10]王山海.基于深度学习神经网络的语音识别研究[D].桂林:桂林电子科技大学,2015.
[11]MOHAMED A R,DAHL G E,HINTON G E.Acoustic Modeling using deep belief networks[J].Audio,Speech,and Language Processing,2012,20(1):14-22.
[12]SHARMA H V,HASEGAWA-JOHNSON M.State-transition interpolation and MAP adaptation for HMM-based dysarthric speech recognition[C]//Proceedings of the NAACL HLT 2010 Workshop on Speech and Language Processing for Assistive Technologies.Association for Computational Linguistics,2010:72-79.
[2]李建文,李岩.多通道皮肤听声器质量检测系统的研究[J].计算机测量与控制,2013,21(11):2941-2943.
[3]SHANNON C.Prediction and entropy of printed English[J].In Bell Technical Journal,1951,30:50-64.
[4]屈丹,张文林.基于本征音子说话人子空间的说话人自适应算法[J].电子与信息学报,2015,37(6):1350-1356.
[5]Gri FFITHS V,KALDI S,THOMPSON S.Mid-career teacher educators:comparing professional learning and academic content in England and Greece[J].Separation&Purification Technology,2010,39(3):181-188.
[6]POVEY D,GHOSHAL A,BOULIANNE G,et al.The kaldi speech recognition toolkit[C]//Proc of IEEE 2011 Workshop on Automatic Speech Recognition and Understanding,Waikoloa Hawaii,2011:128-131.
[7]谢怡宁,黄金杰,何勇军.噪声环境下智能机器人语音控制特征提取方法[J].北京邮电大学学报,2013,36(3):83-87.
[8]张建华.基于深度学习的语音识别应用研究[D].北京:北京邮电大学,2015.
[9]陆俊,张琼,杨俊安,等.嵌入深度信念网络的点过程模型用于关键词检出[J].信号处理,2013,7(12):865-872.
[10]王山海.基于深度学习神经网络的语音识别研究[D].桂林:桂林电子科技大学,2015.
[11]MOHAMED A R,DAHL G E,HINTON G E.Acoustic Modeling using deep belief networks[J].Audio,Speech,and Language Processing,2012,20(1):14-22.
[12]SHARMA H V,HASEGAWA-JOHNSON M.State-transition interpolation and MAP adaptation for HMM-based dysarthric speech recognition[C]//Proceedings of the NAACL HLT 2010 Workshop on Speech and Language Processing for Assistive Technologies.Association for Computational Linguistics,2010:72-79.