公开(公告)号：JP2010009446A

公开(公告)日：2010-01-14

申请号：JP2008170021

申请日：2008-06-30

Applicant: Internatl Business Mach Corp ,  インターナショナル・ビジネス・マシーンズ・コーポレーションInternational Business Maschines Corporation

Inventor： ITO NOBUYASU ,  KURATA TAKEHITO

IPC: G06F17/30 ,  G06F3/16 ,  G10L15/00 ,  G10L15/18

Abstract: PROBLEM TO BE SOLVED: To facilitate registration of a new word and input of a keyword without being conscious of the contents of a voice recognition dictionary as far as possible in order to connect voice recognition to succeeding language processing or retrieval. SOLUTION: In the registration of a new word or the input of a keyword, "reading" is input by a user at first. The reading is converted from pronunciation into notation by the same language model as a language model for voice recognition, and thereby a Kana-Kanji notation is obtained. Then, the obtained Kana-Kanji notation is properly compared with a corrected word and an original character string to identify the unknown word of the voice recognition dictionary. A converted keyword can be used for retrieving retrieval data formed by voice recognition of a voice file. The unknown word portion can be properly registered in the voice recognition dictionary. COPYRIGHT: (C)2010,JPO&INPIT

Abstract translation: 要解决的问题：为了使语音识别与后续语言处理或检索连接，尽可能地尽可能地意识到语音识别字典的内容，便于注册新词和关键字的输入。

解决方案：在注册新词或关键字的输入时，用户首先输入“阅读”。 通过与语音识别的语言模型相同的语言模型将阅读从发音转换成符号，从而获得假名 - 汉字符号。 然后，将获得的假名 - 汉字符号与正确的词和原始字符串进行适当比较，以识别语音识别词典的未知词。 转换的关键字可用于检索由语音文件的语音识别形成的检索数据。 未知字部分可以正确地注册在语音识别字典中。 版权所有（C）2010，JPO＆INPIT

公开(公告)号：JP2008164647A

公开(公告)日：2008-07-17

申请号：JP2006350508

申请日：2006-12-26

Applicant: Internatl Business Mach Corp ,  インターナショナル・ビジネス・マシーンズ・コーポレーションInternational Business Maschines Corporation

Inventor： ITO NOBUYASU ,  KURATA TAKEHITO

IPC: G10L15/18 ,  G10L15/00 ,  G10L15/04 ,  G10L15/10

CPC classification number: G10L15/04 ,  G10L15/19

Abstract: PROBLEM TO BE SOLVED: To split interactive voice into utterance units by using recognition response.
SOLUTION: An apparatus for splitting interactive voice into the utterance units is disclosed and the apparatus has: a word data base stored with description and pronunciation of a word; a grammar data base stored with grammar including connection information between words; a pause detection section for detecting a position of pause in a channel in which main utterance is performed regarding the interactive voice which has been input by at least two channels; a detection section for detecting a position of the recognition response of channels in which the main utterance is not performed; a border candidate extraction section for extracting a border candidate of the main utterance by extracting the pause which is present in a specified period before and after a position of the recognition response as a base point; and a recognition section for dividing the utterance split by the extracted border candidate into an optimum utterance unit by referring to the word data base and the grammar data base to output a word string.
COPYRIGHT: (C)2008,JPO&INPIT

Abstract translation: 要解决的问题：通过使用识别响应将交互式语音分割成话语单元。 公开了一种用于将交互式语音分解成发声单元的装置，该装置具有：存储有单词的描述和发音的单词数据库; 一种语法数据库存储与语法包括词之间的连接信息; 暂停检测部分，用于检测在至少两个通道已经输入的交互式语音中执行主要发声的频道中的暂停位置; 用于检测不执行主要发音的频道的识别响应的位置的检测部分; 边界候选提取部分，通过提取在识别响应的位置之前和之后的指定时段中存在的暂停作为基点来提取主要话语的边界候选; 以及识别部分，用于通过参考字数据库和语法数据库将提取的边界候选者的话语分割划分成最佳发声单元，以输出字串。 版权所有（C）2008，JPO＆INPIT

公开(公告)号：JP2002041080A

公开(公告)日：2002-02-08

申请号：JP2000210599

申请日：2000-07-11

Applicant: IBM

Inventor： MORI SHINSUKE ,  NISHIMURA MASAFUMI ,  ITO NOBUYASU

IPC: G06F17/28 ,  G06F40/00 ,  G10L15/18

Abstract: PROBLEM TO BE SOLVED: To provide a vocabulary prediction method which can raise the accu racy of prediction. SOLUTION: Using string of a partial parse tree covering the word string to the time in each step of word prospect, the partial parse tree having a useful information is selected by the prospect of the next word. In other word, the accuracy of the prospect is improved by selecting the most useful word and/or word string for the prospect of the next word based upon the relation structure and the word string to be a history. Specifying the partial parse tree relating to a word to be prospected, thereafter, the next word is prospected from the partial parse tree, that is, the word and/or word string estimated to be a relation having the connection with the word of prospective object.

公开(公告)号：JPH07319900A

公开(公告)日：1995-12-08

申请号：JP10818694

申请日：1994-05-23

Applicant: IBM JAPAN

Inventor： ITO NOBUYASU

IPC: G06F17/30

Abstract: PURPOSE:To balance a necessary spatial cost and a retrieval cost by deciding a prefix part from a partial string where a small number of input characters are decided and retrieving 'TRIE' again. CONSTITUTION:A candidate character lattice is inputted from an input device 307 and is stored in a prescribed area on the main storage of a computer as a candidate character lattice 308 by a candidate character lattice storage means 302 with the control of an input means 301. The storage means 302 supplies the candidate character lattice 308 to reference or transfers it. The input means 301 stores the specified content of a file in a magnetic disk into the candidate character lattice 308 through the input device 307. A retrieval work quantity estimation means 303 calculates a retrieval start position from which the work quantity of dictionary retrieval can be expected to be small from the candidate character lattice and data 309 of the number of average branches, which is obtained at the time of generating a TRIE dictionary and is preserved in the magnetic disk.

公开(公告)号：JPH06111075A

公开(公告)日：1994-04-22

申请号：JP23372492

申请日：1992-09-01

Applicant: IBM

Inventor： ITO NOBUYASU ,  MARUYAMA HIROSHI

IPC: G06K9/72

Abstract: PURPOSE: To provide a method which can evaluate the result of postprocessing of character recognition of a Japanese sentence by itself and inform an operator of the evaluation result. CONSTITUTION: Paths which are grammatically established are searched for on the basis of the result of the character recognition and the restrictions of Japanese, the costs accompanying the paths are calculated, and on the basis of the costs, candidate paths are selected. Then a conviction degree detecting means 7 finds the degree of conviction as to a candidate characters (REN) of a specific column the optimum candidate path with the best cost passes through from the cost accompanying the optimum candidate path and the cost accompanying a candidate path passing through a candidate character (UN) other than the candidate character, and replaces the candidate character or warns the operator according to the degree of conviction.

公开(公告)号：JPH02250188A

公开(公告)日：1990-10-05

申请号：JP7044289

申请日：1989-03-24

Applicant: IBM JAPAN

Inventor： ITO NOBUYASU

IPC: G06K9/70 ,  G06F17/30 ,  G06K9/62 ,  G06T7/00 ,  G10L11/00 ,  G10L15/06 ,  G10L15/12

Abstract: PURPOSE:To reduce the duplicate recursion formula calculation of templates having the same partial label sequence from the first by making plural templates into a tree structure dictionary corresponding to each access bus. CONSTITUTION:A tree structure dictionary 11 where each label corresponds to one node is generated from plural label sequences forming templates and is held in a storage device 12. A buffer area is reserved in the storage device with respect to the node of each label of the tree structure dictionary 11. Node selection 13 of the tree structure dictionary 11 is performed in the depthwise direction, and stage calculation execution 14 of the input label sequence is performed with respect to the label corresponding to the selected node, and the result is held in the buffer reserved for the selected node, and this operation is repeated. Thus, the duplicate recursion formula calculation of templates having the same partial label sequence from the first is reduced.

公开(公告)号：JPS63198154A

公开(公告)日：1988-08-16

申请号：JP2370687

申请日：1987-02-05

Applicant: IBM

Inventor： ITO NOBUYASU ,  TAKAHASHI HIROYASU

IPC: G06F17/21 ,  G06F17/24 ,  G06F17/27

公开(公告)号：JP2009014888A

公开(公告)日：2009-01-22

申请号：JP2007174862

申请日：2007-07-03

Applicant: Internatl Business Mach Corp ,  インターナショナル・ビジネス・マシーンズ・コーポレーションInternational Business Maschines Corporation

Inventor： TAKEUCHI HIROYOSHI ,  NEGISHI SHIHO ,  ITO NOBUYASU

IPC: G10L15/10 ,  G10L15/00

CPC classification number: G10L15/26 ,  G10L2015/088

Abstract: PROBLEM TO BE SOLVED: To provide an interactive processing device, an interaction processing method and a computer program, capable of extracting a necessary uttering section in a specified field from a conversation data, without requiring prior knowledge regarding the data and the application field. SOLUTION: The interactive processing device 1 comprises a processing object data extracting section 11 for extracting a plurality of processing object data, including a pattern adaptation section which is adapted to an utterance pattern that is an utterance structure, derived from a content of general conversation which does not depend on the field that is input by an utterance pattern input section 32, from among a plurality of utterance data in which a plurality of conversation contents, regarding one field that is input by an utterance data input section 31; a feature amount extracting section 12 for extracting feature amount which is common for the plurality of pattern adaptation sections by taking each of the pattern adaptation section from the plurality of extracted processing object data; and an essential data extracting section 15 for extracting a necessary data in one field which is included in the plurality of utterance data by using the extracted feature amount. COPYRIGHT: (C)2009,JPO&INPIT

Abstract translation: 要解决的问题：提供一种交互式处理装置，交互处理方法和计算机程序，能够从会话数据中提取特定领域中的必要的发声部分，而不需要关于数据和应用的事先知识 领域。 交互处理装置1包括：处理对象数据提取部11，用于提取多个处理对象数据，该处理对象数据包括适合于作为话语结构的发音模式的模式适配部， 关于由话音数据输入部31输入的一个场的多个话音内容的多个话音数据中，不依赖于由话音模式输入部32输入的场的一般对话; 特征量提取部分12，用于通过从多个提取的处理对象数据中取出模式自适应部分中的每一个，提取多个模式自适应部分共同的特征量; 以及必要数据提取部分15，用于通过使用提取的特征量来提取包含在多个话语数据中的一个字段中的必要数据。 版权所有（C）2009，JPO＆INPIT

公开(公告)号：JP2002258890A

公开(公告)日：2002-09-11

申请号：JP2001044186

申请日：2001-02-20

Applicant: IBM

Inventor： ITO NOBUYASU ,  NISHIMURA MASAFUMI

IPC: G06F3/16 ,  G10L15/02 ,  G10L15/18

Abstract: PROBLEM TO BE SOLVED: To provide a model-setting function which is more suitable, relating to the removal of redundant word from the information obtained in user assignment or speaker registration. SOLUTION: A redundant word language model 322, specified as redundant words, is provided in addition to a versatile language model 320 as a language model 32a. The automatic deletion of the redundant words can be made, by using this redundant word language model 322.