Data processing: speech signal processing – linguistics – language – Speech signal processing – Recognition
Reexamination Certificate
2001-03-13
2004-10-12
Chawan, Vijay (Department: 2655)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Recognition
C704S270100, C704S273000
Reexamination Certificate
active
06804647
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to the field of speech recognition. In particular the present invention relates to a system and method for on-line unsupervised adaptation in speaker verification.
BACKGROUND OF THE INVENTION
Natural language speaker verification systems are currently in use for responding to various forms of commerce via a telephone network. One example of such a system is utilized in conjunction with a stock brokerage. According to this system, once a caller's voice has been authenticated, the caller may obtain a quotation for the price of a particular stock issue, purchase or sell a particular number of shares at market price or a predetermined target price among other types of transactions. Natural language systems can also be used to respond to such things as requests for telephone directory assistance.
One of the most significant sources of performance degradation in a speaker verification system is the acoustic mismatch between the enrollment and subsequent verification sessions. Acoustic mismatches may occur as a result of differences in transducers, acoustic environment, and communication channel characteristics (e.g., varying channels associated with combinations of different subnetworks utilized in a telephone call). Of the factors contributing to acoustic mismatch in telephony applications, it has been shown that the mismatch in transducers of telephone handsets is the most dominant source of performance degradation.
To address the acoustic mismatch problem, a variety of approaches for robust speaker recognition have been developed in the past several years. These approaches include robust feature, model, and score-based normalization techniques. These approaches use off-line development data to compensate for the effects of acoustic mismatch that will be present when the system is used on-line.
Another approach has been developed that uses on-line unsupervised adaptation to “learn” the unseen channel characteristics automatically while the system is being used in the field. Unsupervised systems do not require human intervention during the verification process. Compared to off-line adaptation approaches, on-line approaches provides significantly more data for parameter estimation than typically available to the speaker verification system, facilitating more sophisticated modeling approaches and automated parameter tuning. Furthermore, rather than predicting the effects of acoustic mismatch with development data, the effects can be observed directly from this additional data.
Prior approaches to on-line unsupervised adaptation suffered from numerous limitations. For example, adaptation of the speaker model suffered negative effects from impostor attacks, it significantly increased the size of the speaker model, and it degraded the performance on the enrollment handset-type when adapting on new handset types.
SUMMARY OF THE INVENTION
The present invention introduces a system and method for unsupervised, on-line, adaptation in speaker verification. In one embodiment, a method for adapting a speaker model to improve the verification of a speaker's voice, comprises detecting a channel of a verification utterance; learning vocal characteristics of the speaker on the detected channel; and transforming the learned vocal characteristics of the speaker from the detected channel to the speaker model of a second channel.
Other features of the present invention will be apparent from the accompanying drawings and from the detailed description, which follows.
REFERENCES:
patent: 5528731 (1996-06-01), Sachs et al.
patent: 5774841 (1998-06-01), Salazar et al.
patent: 5950157 (1999-09-01), Heck et al.
patent: 5960397 (1999-09-01), Rahim
patent: 6032115 (2000-02-01), Kanazawa et al.
patent: 6233556 (2001-05-01), Teunen et al.
patent: 6266633 (2001-07-01), Higgins et al.
patent: 6327565 (2001-12-01), Kuhn et al.
patent: 2002/0077828 (2002-06-01), Robbins
patent: 0424071 (1991-04-01), None
L.P. Heck and N. Mirghafori, “On-line Unsupervised Adaptation in Speaker Verification,” Proceedings of the International Conference on Spoken Language Processing, pp. 1-4, Beijing, China, Oct. 18, 2000.
Douglas A. Reynolds, Automatic Speaker Recognition Using Gaussian Mixture Speaker Models, vol. 8, No. 2, 1995, The Lincoln Laboratory Journal, pp. 173-192.
Remco Teunen, Ben Shahshahani, Larry Heck, “A Model-Based Transformational Approach To Robust Speaker Recognition,” Nuance Communication, 1380 Willow Rd, Menlo Park, CA 94025, USA.
C. Fredouille, J. Mariethoz, C. Joboulet, J. Hennebert, J.-F. Bonastre, C. Mokbel, F. Bimbot, “Behavior Of A Bayesian Adaptation Method For Incremental Enrollment In Speaker Verification,” ICASSP2000 Istanbal, Turkey.
Owen Kimbal, Michael Schmidt, Herbert Gish, Jason Waterman, “Speaker Verification With Limited Enrollment Data,” BBN Systems & Tech., 70 Fawcett St., Cambridge, MA 02138 USA, Eurospeech 97, Rhodes, Greece.
Aaron E. Rosenberg, Chin-Hui Lee, Frank K. Soong, “Sub-Word Unit Talker Verification Using Hidden Marker Models” IEEE ICASSP 90.
William Mistretta, Kevin Farrell, “Model Adaptation Methods For Speaker Verification,” T-Netix/SpeakEZ Inc., 67 Inverness Drive East Englewood, CO 80112, ICASSP98 Seattle, WA.
Tatsuo Matsuoka, Chin-Hui Lee, “A Study Of On-Line Bayesian Adaptation For HMM-Based Speech Recognition,” Speech Research Dept., AT&T Bell Lab., Murray Hill, N.J. 07974, USA, Eurospeech 93 vol. 2. Verlin, Germany.
Heck Larry Paul
Mirghafori N. Nikki
Blakely , Sokoloff, Taylor & Zafman LLP
Chawan Vijay
Nuance Communications
Opsasnick Michael N.
LandOfFree
Method and system for on-line unsupervised adaptation in... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and system for on-line unsupervised adaptation in..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and system for on-line unsupervised adaptation in... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3285345