专利快速检索-快速检索全球专利，免费商用专利数据库-IPRDB

1. 发明专利

JPH07115370A HIGH SPEED VECTOR QUANTIZER AND HIGH SPEED VECTOR QUANTIZING METHOD 失效
公开(公告)号：JPH07115370A
公开(公告)日：1995-05-02
申请号：JP28592193
申请日：1993-10-18
申请人： NIPPON DENSO CO
发明人： SATO TATSUYA , AKAHORI ICHIRO , FURUHATA KEIJI , SATO HIDEKI
IPC分类号： H03M7/30
摘要： PURPOSE:To provide the vector quantizer and the vector quantizing method, by which a code vector of the minimum distortion is obtained, and the data processing capacity is improved. CONSTITUTION:The figure shows a block diagram of the high speed vector quantizer, consists of a retrieval control means 1b having a distance calculating circuit 1g of an exclusive IC, and a code book(CB) 1a, and the CB 1a is a ROM having a node dividing condition of a bisected tree structure up to an m-th stage. By using a reduced multi-dimensional tree, the divided area is reduced more remarkably. To which vector of the CB 1a an input vector 1h is the nearest is compared with a square value of a distance by a infinity norm to an area formed by each node, after following a bisected tree and deriving a temporary solution, all the nodes are checked, and in the end, a result if outputted from a back track control part 1c. A calculation of the infinity norm distance is only subtraction and a square operation, a square root calculation containing a floating point is not required, and the calculation can be saved considerably, compared with deriving a Euclidean distance and executing the comparison.

2. 发明专利

JPH0798598A SOUND RECOGNIZING DEVICE 失效
公开(公告)号：JPH0798598A
公开(公告)日：1995-04-11
申请号：JP24151793
申请日：1993-09-28
申请人： NIPPON DENSO CO
发明人： FURUHATA KEIJI , AKAHORI ICHIRO , SATO TATSUYA
IPC分类号： G10L15/14 , G10L3/00
摘要： PURPOSE:To provide a sound recognizing device which can reduce memory capacity, for example, a device which can be effectively applied to a sound dial telephone and the like. CONSTITUTION:An input voice is inputted to a feature sampling section 11, and time series of feature parameters is outputted. In a voice learning mode, an output from the feature sampling section 11 is sent to a parameter deciding section 13 through a switch 12, parameters of each registered word to be recognized are decided, similarity between parameters is calculated while similar parameters are integrated in a compression section. And this compressed HMM parameter is held by a holding section 15. At the time of recognition, an output from the feature sampling section 11 is sent to a likelihood calculating section 16 through the switch 12, likelihood of each word is calculated in accordance with the stored HMM parameter, and a word having the highest likelihood in a word deciding section is outputted as a recognizing word.

3. 发明专利

JPH08129396A METHOD AND DEVICE FOR VOICE RECOGNITION 失效
公开(公告)号：JPH08129396A
公开(公告)日：1996-05-21
申请号：JP26868094
申请日：1994-11-01
申请人： NIPPON DENSO CO
发明人： FURUHATA KEIJI
IPC分类号： G10L15/10 , G10L15/02 , G10L15/08 , G10L15/14 , G10L15/22 , G10L3/00 , G10L9/10
摘要： PURPOSE: To improve the using convenience by inputting an arbitrary voice selected by a speaker at the time of performing a speaker adaptation. CONSTITUTION: A feature extracting section 1 acoustically analyzes a word voice generated by a user and extracts the time series of feature parameters. A recognition section 3 performs liklihood calculations of each recognition parameter against the feature parameters based on the time series of the extracted feature parameters and the recognition parameter group stored in a recognition parameters holding section 2. Plural recognition object words having a higher computed liklihood are selected as the word group which is assumed to be corresponding to the input voices and are displayed on a result display section 4. A selection result reception section 5 outputs signals which correspond the words selected by the user from the display contents of the section 4 and the time series of the feature parameters from the section 1. A speaker adaptiveness section 6 performs a speaker adaptation in which the recognition parameters are updated so as to match the parameters with the user's voice quality based on the output signals from the section 5.

4. 发明专利

JPH0927845A AUTOMATIC DIAL TELEPHONE SET 失效
公开(公告)号：JPH0927845A
公开(公告)日：1997-01-28
申请号：JP17750495
申请日：1995-07-13
申请人： NIPPON DENSO CO
发明人： FURUHATA KEIJI , SHIBUYA SEIICHI
IPC分类号： H04M1/27
摘要： PROBLEM TO BE SOLVED: To obtain an automatic dial telephone set in which a telephone number is entered in voice. SOLUTION: A telephone number entered by the voice of the user is recognized by a voice recognition section 11, a numeral string recognized is used for voice information by a voice synthesis section 13 and it is recognized by the user. The voice input is recognized as a voice by each block of 2 to 5 digits and confirmed through voice synthesis. The user depresses a switch corresponding to the start of entry and confirmation of the synthesized voice, it is recognized by a switch input reception section 12, and a command is given to the voice recognition section 11 to command voice recognition. Furthermore, a command is given to a memory section 14 and an entered numeral string confirmed to be correct is stored. When the entry of all the numbers is confirmed, the stored number string in the memory section 14 is sent to a dial section 15, and a telephone circuit section 16 sends it to a telephone line 17 as a dial signal.

5. 发明专利

JPH0772892A DEVICE FOR VOICE RECOGNITION 失效
公开(公告)号：JPH0772892A
公开(公告)日：1995-03-17
申请号：JP21840993
申请日：1993-09-02
申请人： NIPPON DENSO CO
发明人： AKAHORI ICHIRO , FURUHATA KEIJI , SATO TATSUYA
IPC分类号： G10L15/10 , G10L15/14 , G10L3/00
摘要： PURPOSE:To provide a voice recognition device furthermore improving a word recognition rate. CONSTITUTION:A sound input is supplied to a characteristic extraction part 13 together with a sound section detection part 11, and the time sequence of a characteristic parameter of an input sound section is extracted, and the similarity is found by a similarity adding part 13 based on a madel stored in an HMM storage part 14. To the similarity obtained in such a manner, a correction amount dealing with respective words stored in a correction amount storage part 16 is added in a correction adding part 15, and the maximum similarity of the corrected similarity is decided by a word decision part 17, and the recognition result of word is obtained.

你已经成功收藏专利！

检索式保存成功!

IPRDB

热门服务

关于我们

友情链接

联系方式