公开(公告)号：EP0312209A2

公开(公告)日：1989-04-19

申请号：EP88308585.4

申请日：1988-09-16

申请人： International Business Machines Corporation

发明人： Nishimura, Masafumi Denendomiiru 405

IPC分类号： G10L5/06

CPC分类号： G10L15/14

摘要： The present invention relates to a speech recognition system including Markov models which are adapted to be trained by an initial training label set and initial training speech using adaptation speech, and are used to recognise input speech. According to the invention, the system is characterised in that it comprises means for interpreting the adaptation speech into adaptation label strings using an adaptation label set; means for connecting each label in each of the adaptation label strings with each state or each state transition of a Markov model which corresponds to the adaptation label strings concerned; means for determining the confusion probability of each label in the initial training label set and each label in the adaptation label set being confused with each other, based on the connection between each label in the adaptation label set and each of the states or state transitions, and the parameter values of the Markov model concerned with the initial training set; and means for determining parameter values of each of the Markov models with respect to the adaptation label set, based on the confusion probabilities and the parameter values of the Markov model concerned with the initial label set.

摘要翻译： 本发明涉及一种包括马可夫模型的语音识别系统，该模型适于由初始训练标签组和使用自适应语音的初始训练语音进行训练，并用于识别输入语音。 根据本发明，该系统的特征在于它包括使用自适应标签集将适应语音解释成自适应标签串的装置; 用于将每个适配标签​​串中的每个标签与对应于相关适应标签串的马尔可夫模型的每个状态或每个状态转换连接的装置; 用于基于所述自适应标签集合中的每个标签与所述状态或状态转换中的每一个之间的连接来确定所述初始训练标签集中的每个标签和所述自适应标签集中的每个标签彼此混淆的每个标签的混淆概率的装置， 以及与初始训练集相关的马尔科夫模型的参数值; 以及用于基于与初始标签集相关的马尔科夫模型的混淆概率和参数值来确定关于自适应标签集的每个马尔科夫模型的参数值的装置。

公开(公告)号：EP0312209A3

公开(公告)日：1989-08-30

申请号：EP88308585.4

申请日：1988-09-16

申请人： International Business Machines Corporation

发明人： Nishimura, Masafumi Denendomiiru 405

IPC分类号： G10L5/06

CPC分类号： G10L15/14

摘要： The present invention relates to a speech recognition system including Markov models which are adapted to be trained by an initial training label set and initial training speech using adaptation speech, and are used to recognise input speech. According to the invention, the system is characterised in that it comprises means for interpreting the adaptation speech into adaptation label strings using an adaptation label set; means for connecting each label in each of the adaptation label strings with each state or each state transition of a Markov model which corresponds to the adaptation label strings concerned; means for determining the confusion probability of each label in the initial training label set and each label in the adaptation label set being confused with each other, based on the connection between each label in the adaptation label set and each of the states or state transitions, and the parameter values of the Markov model concerned with the initial training set; and means for determining parameter values of each of the Markov models with respect to the adaptation label set, based on the confusion probabilities and the parameter values of the Markov model concerned with the initial label set.

公开(公告)号：EP0312209B1

公开(公告)日：1992-11-25

申请号：EP88308585.4

申请日：1988-09-16

申请人： International Business Machines Corporation

发明人： Nishimura, Masafumi Denendomiiru 405

IPC分类号： G10L5/06

CPC分类号： G10L15/14

摘要： The present invention relates to a speech recognition system including Markov models which are adapted to be trained by an initial training label set and initial training speech using adaptation speech, and are used to recognise input speech. According to the invention, the system is characterised in that it comprises means for interpreting the adaptation speech into adaptation label strings using an adaptation label set; means for connecting each label in each of the adaptation label strings with each state or each state transition of a Markov model which corresponds to the adaptation label strings concerned; means for determining the confusion probability of each label in the initial training label set and each label in the adaptation label set being confused with each other, based on the connection between each label in the adaptation label set and each of the states or state transitions, and the parameter values of the Markov model concerned with the initial training set; and means for determining parameter values of each of the Markov models with respect to the adaptation label set, based on the confusion probabilities and the parameter values of the Markov model concerned with the initial label set.