Data processing: speech signal processing – linguistics – language – Modification of at least one characteristic of speech waves – Transformation of speech into a nonaudible representation,...
Patent
1997-08-25
2000-06-27
Teska, Kevin J.
Data processing: speech signal processing, linguistics, language
Modification of at least one characteristic of speech waves
Transformation of speech into a nonaudible representation,...
704240, 704250, 704255, G10L 506
Patent
active
060816600
DESCRIPTION:
BRIEF SUMMARY
TECHNICAL FIELD
This invention relates to a method for forming a cohort for use in identification of an individual, and to a method of identification of an individual on the basis of that cohort. The method is concerned primarily, but not exclusively, with forming a cohort for use in identification of individuals on the basis of the degree of conformity of characteristics of voice sounds, but may be applied to identification on the basis of other characteristics of individuals.
BACKGROUND OF THE INVENTION
In determining whether an individual is or is not a particular pre-identified individual ie a "client", comparison may be made as between pre-determined parameters relating to the pre-determined person and those measured when any individual is presented for verification. Particular parameters which may be used include parameters relating to speech, although parameters relating to other characteristics may be used. Among those other characteristics are parameters relating to how the presenting individual writes, uses a computer mouse, or uses a computer or other keyboard.
One method of identification, or verification, of whether or not an individual presenting for verification is or is not a pre-determined individual makes use of client models representing each of a population of individuals. Characteristics relating to a person presenting for verification are measured and compared with the characteristics for one or more of the total population. If the characteristics for the person presenting for verification match those for a particular one of the population, then the verification system makes a determination that the presenting person is the particular individual for which the characteristics match. A difficulty with systems of this kind is that values for characteristics for any person presenting may differ from reference values for that person which are used by the system. For example, the values for characteristics used by the system would normally comprise stored values measured in a previous test on the individual, the stored value then being compared with those measured when the person presents for verification. However, naturally occurring variations may exist as between those values stored and those which arise when a verification procedure is carried out. In the case of verification on the basis of characteristics relating to utterances of a person, those variations may, for example, comprise phonetic variations, variations due to environmental conditions and intra speaker variations. Thus, a person may utter a vowel in one fashion when the vowel appears in one word, and in a different fashion when it appears in another word. Again, the test conditions under which the original characteristic values were determined may be noise free, but there may be noise present in the environment when the individual presents for verification. Generally, the, it is not surely possible to effect identification simply on the basis of direct equatability of measured characteristics with those stored for the individual in question. Normally, comparison is effected as between characteristic values for more than one of the population, the determination of identity being made on the basis of the "distance" between the characteristics as stored for more than one of the population and those measured at verification. The characteristics which are measured in the verification process may be multi dimensional. For example, it has been found convenient to use cepstral analysis techniques to analyse the speech of a population and the person presenting for verification. Overlapping samples of, say, 30 millisecond may be taken of the amplitude-time wave form recorded during speech. In this case, it is convenient to generate 15 cepstral coefficients and to generate a model representing each member of the population and of the person presenting for verification, the models being 15 dimensional and with, for example, 128 points. The set of such points is commonly referred to as a code book for the person in question.
In the
REFERENCES:
patent: 4897878 (1990-01-01), Boll et al.
patent: 4910782 (1990-03-01), Watari
patent: 5271088 (1993-12-01), Bahler
patent: 5303299 (1994-04-01), Hunt et al.
patent: 5307444 (1994-04-01), Tsuboka
patent: 5414755 (1995-05-01), Bahler et al.
patent: 5548647 (1996-08-01), Naik et al.
patent: 5864810 (1999-01-01), Digalakis et al.
Chen, Fangxin, et al., "Hybrid Threshold Approach In Text-Independent Speaker Verification," Research School of Information Sciences and Engineering. Australian National University, pp. 79-82.
Millar, J. Bruce, et al. "Overview of Speaker Verification Studies Towards Technology For Robust User-Conscious Secure Transactions," Research School of Information Sciences and Engineering, Australian National University.
Millar, J. Bruce, et al. "The Efficacy of Cohort Normalisation In A Speaker Verification Task Under Different Types of Speech Signal Variance," Research School of Information Sciences and Engineering, Australian National University.
Hawkins, Simon, et al. "Modelling Individual Speaker Characteristics By Descriving A Speaker's Vowel Distribution In Articulatory, Cepstral and Formant Space," Computer Sciences Laboratory, Research School of Information Sciences and Engineering, Australian National University, pp. 342-347.
Rosenberg, Aaron E., et al. "The Use of Cohort Normalized Scores For Speaker Verification," Speech Research Department, AT&T Bell Laboratories, pp. 599-602.
Davis, Steven B. and Paul Mermelstein, "Comparison Of Parametric Representations For Monosyllabic Word Recognition In Continuously Spoken Sentences," IEEE Transactions On Acoustics, Speech and Signal Processing, ASSP-28, pp. 357-366, 1980.
X. Yang, J. Millar and I. Macleod, On the Sources of Inter- & Intra-Speaker Variability in the Acoustic Dynamics of Speech, IEEE, 4th Int'l Conf. on Spoken Language, 1996, ICSLP '96, vol. 3, pp. 1792-1795.
P. Mousty et al., The Role of Neighborhood Relative Frequency in Spoken Word Recognition, IEEE, 4th Int'l Conf. on Spoken Language, 1996, ICSLP '96, vol. 4, pp. 2498-2501.
Chen Fangxin
Laverty William
Macleod Iain Donald Graham
Millar John Bruce
Frejd Russell W.
Teska Kevin J.
The Australian National University
LandOfFree
Method for forming a cohort for use in identification of an indi 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 for forming a cohort for use in identification of an indi, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for forming a cohort for use in identification of an indi will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-1791321