发明名称 SOUND RECOGNIZING SYSTEM
摘要 PURPOSE:To heighten a reconization rate of a short word, to reduce the number of candidates and to improve a recognizing speed by operating a recognition in phrase unit and recognizing the phrase unit having the same one in a structural condition. CONSTITUTION:When the recognition of the phrase of a previous stage is carried out for the part of speech of a conceptual word at the start of a sentence at the initial stage of a recognizing process, the part of speech of a conceptual word next to a candidate gained at a previous stage is decided by a structural condition change network (1). The phrase is cut off from an input sound element string based on the length of the sound element string of the conceptual word of the gained part of speech (2). The degrees of similarities among sound element strings in the dictionary of the phrase and the conceptual word are gained and the one of a set value is made a candidate (3). Regarding the part of speech of the conceptual word, the part of speech of a following functional word in the structural condition change network is decided (4) and the sound element string of the phrase is made from the conceptual word candidate and the functional word (5). A phrase is cut off from the input sound element string (6), the degree of similarity between the phrase and the made phrase is gained (7) and when a phrase candidate is left and the phrase of the previous stage is used as candidate, the squaring of a set number is gained.
申请公布号 JPH0695684(A) 申请公布日期 1994.04.08
申请号 JP19920247453 申请日期 1992.09.17
申请人 MEIDENSHA CORP 发明人 SEKIGUCHI YOSHIHIRO;MATSUURA YOSHIHIRO
分类号 G10L15/04;G10L15/10;G10L15/18;(IPC1-7):G10L3/00;G10L3/00 主分类号 G10L15/04
代理机构 代理人
主权项
地址