-
公开(公告)号：JPH1173195A
公开(公告)日：1999-03-16
申请号：JP20250898
申请日：1998-07-17
Applicant: LUCENT TECHNOLOGIES INC
Inventor： JUANG BIING-HWANG , LEE CHIN-HUI , LI QI P , ZHOU QIRU
Abstract: PROBLEM TO BE SOLVED: To eliminate the need for enormous investment of the time and labor for a training process by adopting a constitution in which one speech model series is determined in accordance with the partial word transcription of a word series and the partial word transcription consists of a partial word series. SOLUTION: The characteristic of spoken speech utterance is compared with one speech model series and the reliability level reflecting the word series consisting of >=1 words associated with an individual having proposed identification is determined. The one speech model series among the speech model series corresponds to the speech reflecting the word series associated with the individual having the proposed identification is determined. The one speech model series is determined in accordance with the partial word transcription of the word series. The partial word transcription consists of the partial word series consisting of >=1 partial words. A reliability measure module 34 judges the reliability measure, at which a password phrase associated with the individual having the proposed identification is actually the phrase of test utterance by using the strings of target likelihood evaluation points and the strings of anti- likelihood evaluation points.
-
公开(公告)号：JPH1173196A
公开(公告)日：1999-03-16
申请号：JP20250998
申请日：1998-07-17
Applicant: LUCENT TECHNOLOGIES INC
Inventor： JUANG BIING-HWANG , LEE CHIN-HUI , LI QI P , ZHOU QIRU
Abstract: PROBLEM TO BE SOLVED: To eliminate the need for enormous investment of the time and labor for a training process, by adopting a constitution, in which one speech model series among speech model series corresponds to the speech reflecting the information associated with an individual having proposed identification. SOLUTION: The characteristic of spoken speech utterance is compared with one speech model series and the reliability level reflecting the information associated with an individual having proposed identification is determined by this spoken speed utterance. The one speech model series among the speech model series corresponds to the speech reflecting the information associated with the individual having the proposed identification. A reliability measure module 34 judges the reliability measure, at which a password phrase associated with the individual having the proposed identification is actually the phrase of test utterance by using the strings of target likelihood evaluation points and the strings of anti-likelihood evaluation points.
-
Patent Agency Ranking
公开(公告)号：CA2239339A1
公开(公告)日：1999-01-18
申请号：CA2239339
申请日：1998-05-29
Applicant: LUCENT TECHNOLOGIES INC
Inventor： LEE CHIN-HUI , JUANG BIING-HWANG , LI QI P , ZHOU QIRU
Abstract: A method and apparatus for authenticating a proffered identity of a speaker in which the verbal information content of a speaker's utterance, rather than the v ocal characteristics of the speaker, are used to identify or verify the identity of a speaker. Specifically, features of a speech utterance spoken by a speaker are compared wi th at least one sequence of speaker-independent speech models, where one of these sequences of speech models corresponds to speech reflecting information associat ed with an individual having said proffered identity. Then, a confidence level that the speech utterance in fact reflects the information associated with the individual having said proffered identity is determined based on said comparison. In accordance wi th one illustrative embodiment, the proffered identity is an identity claimed by th e speaker, and the claimed identity is verified based upon the determined confiden ce level. In accordance with another illustrative embodiment, each of a plurality o f proffered identities is checked in turn to identify the speaker as being a parti cular one of a corresponding plurality of individuals. The features of the speech utteranc e may comprise cepstral (i.e., frequency) domain data, and the speaker-independent spe ech models may comprise Hidden Markov Models of individual phonemes. Since speaker-independent models are employed, the need for each system user to perfor m an individual training session is eliminated.
-
公开(公告)号：DE69800320T2
公开(公告)日：2001-05-10
申请号：DE69800320
申请日：1998-07-07
Applicant: LUCENT TECHNOLOGIES INC
Inventor： JUANG BIING-HWANG , LI QI P , LEE CHIN-HUI , ZHOU QIRU
-
5.发明专利
公开(公告)号：CA2239340A1
公开(公告)日：1999-01-18
申请号：CA2239340
申请日：1998-05-29
Applicant: LUCENT TECHNOLOGIES INC
Inventor： ZHOU QIRU , JUANG BIING-HWANG , LI QI P , LEE CHIN-HUI
Abstract: A method and apparatus for authenticating a proffered identity of a speaker in which the verbal information content of a speaker's utterance, rather than the v ocal characteristics of the speaker, are used to identify or verify the identity of a speaker. Specifically, features of a speech utterance spoken by a speaker are compared wi th at least one sequence of speaker-independent speech models, where one of these sequences of speech models corresponds to speech reflecting information associat ed with an individual having said proffered identity. Then, a confidence level that the speech utterance in fact reflects the information associated with the individual having said proffered identity is determined based on said comparison. In accordance wi th one illustrative embodiment, the proffered identity is an identity claimed by th e speaker, and the claimed identity is verified based upon the determined confiden ce level. In accordance with another illustrative embodiment, each of a plurality o f proffered identities is checked in turn to identify the speaker as being a parti cular one of a corresponding plurality of individuals. The features of the speech utteranc e may comprise cepstral (i. e., frequency) domain data, and the speaker-independent sp eech models may comprise Hidden Markov Models of individual phonemes. Since speaker-independent models are employed, the need for each system user to perfor m an individual training session is eliminated.
-
公开(公告)号：CA2239339C
公开(公告)日：2002-04-16
申请号：CA2239339
申请日：1998-05-29
Applicant: LUCENT TECHNOLOGIES INC
Inventor： LI QI P , ZHOU QIRU , LEE CHIN-HUI , JUANG BIING-HWANG
Abstract: A method and apparatus for authenticating a proffered identity of a speaker in which the verbal information content of a speaker's utterance, rather than t he vocal characteristics of the speaker, are used to identify or verify the identity of a speaker. Specifically, features of a speech utterance spoken by a speaker are compare d with at least one sequence of speaker-independent speech models, where one of these sequences of speech models corresponds to speech reflecting information asso ciated with an individual having said proffered identity. Then, a confidence level that the speech utterance in fact reflects the information associated with the indivi dual having said proffered identity is determined based on said comparison. In accordanc e with one illustrative embodiment, the proffered identity is an identity claimed b y the speaker, and the claimed identity is verified based upon the determined conf idence level. In accordance with another illustrative embodiment, each of a plurali ty of proffered identities is checked in turn to identify the speaker as being a p articular one of a corresponding plurality of individuals. The features of the speech utte rance may comprise cepstral (i.e., frequency) domain data, and the speaker-independent speech models may comprise Hidden Markov Models of individual phonemes. Since speaker-independent models are employed, the need for each system user to pe rform an individual training session is eliminated.
-
公开(公告)号：DE69800320D1
公开(公告)日：2000-10-26
申请号：DE69800320
申请日：1998-07-07
Applicant: LUCENT TECHNOLOGIES INC
Inventor： JUANG BIING-HWANG , LI QI P , LEE CHIN-HUI , ZHOU QIRU
-
-
-
-
-
-
Search Results
7
records
(took 0.024 seconds)
