Speaker verification method

Details Speaker verification method Speaker verification method

1999-04-13

2003-04-29

Horabik, Michael (Department: 2735)

Communications: electrical

Selective

Intelligence comparison for controlling

C340S005850, C340S005800, C340S005810, C340S005220, C340S005520, C704S246000, C704S273000

Reexamination Certificate

active

06556127

ABSTRACT:

The invention relates to a method for enabling a user of a system to obtain access to one or more services according to the preamble of claim
1
.
In many situations, it is necessary to identify users before giving them access to services or systems. These services include remote access to computers, electronic banking transactions, security access control for buildings, etc. In the same way, for many new telephone services it is necessary to verify unambiguously the identity of the caller. The caller identity usually cannot be determined from the access point in the telephone network used by the caller, as a caller may want to access those services from different access points.
An object of the present invention is to provide a method for enabling a user of a system to be identified and thus to obtain access to one or more services in a convenient way.
In a well-known method used for identifying a user in a system, the user is first requested to identify himself by entering a personal identification code (ID). The user is then requested to enter a secret password used to verify the claimed identity. Access to the requested services is given only if the entered password matches the entered ID. This is the method, for example, for using the calling cards proposed by many different telephone network operators. A subscriber who wants to use the service is usually furnished with a card, for example in credit card format, on which a subscriber code or number is printed. Furthermore, the subscriber is given a confidential personal identification code (PIN code) he has to remember. The subscriber code is needed for identifying the subscriber and the PIN code for verifying the entered identity.
To use the service, the subscriber calls a toll-free telephone number which connects him to the server which implements the calling card service. Once the call setup has been established, a speech generator in the server prompts the caller to enter first his subscriber number (which is printed on the card he has previously bought), and then his confidential PIN code. Those numbers may be entered manually on the telephone keyboard or, more comfortably, orally. In this latter case, the digits are recognized by a speech recognizing algorithm implemented in the server.
If the server recognizes the subscriber number entered and if the entered PIN code matches this subscriber number, the subscriber can then be connected to another subscriber within the network. The invoice for the telephone call will be sent automatically to the calling subscriber, independently of the place from which the call was made.
This method is very tedious for the subscriber who has to enter successively two numbers, one of which has to be memorized. The security is not optimal, as very often the PIN code is written down near or on the card.
The U.S. Pat. No. 5,297,194 describes a method in which the subscriber is prompted to enter one unique password digit-by-digit. A speech recognition algorithm running on the digital processing system of the server processes the information received and, using a statistical recognition strategy, determines the digits. After all digits have been recognized, and if the password is valid, the server uses a speaker verification algorithm to determine if the entered password has been spoken by a caller whose voice has been previously assigned to the password. Access to other services is allowed only if the verification algorithm establishes a match; otherwise, the system rejects the access inquiry and the call is terminated.
This method generally requires the subscribers to have previously enrolled in the system. Accordingly, the speaker verification algorithm uses a voice reference database comprising voice references collected from all authorized subscribers. Enrollment is accomplished by having the subscriber repeat a multidigit password several times. A voice reference is then obtained from those utterances. This voice reference is assigned to the password attributed to the subscriber.
This method requires active collaboration of the user at the time of enrollment and thus discourages in particular new subscribers. Moreover enrollment rollment is generally carried out in a single session from a single telephone apparatus, apparatus, for example from the telephone at the address of the new subscriber. The quality of the voice reference registered therefore very much depends upon the characteristics of the telephone apparatus originally used, and the subscriber risks not being recognized if he later calls from a different telephone.
Other speaker recognition methods and algorithms are described by George R. Doddington in “Speaker Recognition—Identifying People by their Voices”, Proceed-ings of the IEEE, vol. 73, No. 11, November 1985, pp. 1651-1664; by John Oglesby in “What's in a number? Moving beyond the equal error rate”, Speech Communication, vol. 17, 1995, pp. 193-208; and by Sadaoki Furui, in “An overview of Speaker Recog-nition Technology”, ESCA Workshop on Automatic Speaker Recognition, Identification and Verification.
The object of the present invention is to provide an improved method for enabling a user of a system to obtain access to one or more services.
This object is attained according to the invention by means of supplementary steps as set forth in the characterizing part of claim
1
.
Preferred embodiments of the invention follow from the dependent claims.
According to the invention, a subscriber receives a subscriber code and a confidential PIN code, as in the prior art. The subscriber code serves to identify the subscriber and the PIN code to verify the claimed identity.
To use the service, the user is first prompted (for example by a speech generator) to enter his subscriber code. This number may, for example, be printed on a subscriber card he has previously bought. A speaker verification algorithm is then used for verifying the identity claimed by the user by making a voice comparison between the user's voice and a voice reference stored in the system and associated with the subscriber code entered by the user. Access to the requested services is given if the user's voice matches said voice reference.
If said voice reference is not good enough to verify the claimed identity with a prescribed degree of reliability, the user is then prompted to enter his confidential PIN code. The PIN code is then checked, and, if the outcome of this check is positive, a new voice reference is created in a background process and associated with the subscriber code entered by the user.
One advantage of the invention, especially if used for identifying callers in a telephone network, is that the voice reference is created and stored during several sessions, usually from different locations and with various telephone apparatus. Thus, the voice reference stored for a given subscriber usually takes very different environmental conditions (noise, telephone lines, microphone, etc.) into account.
According to the invention, a specific enrollment session is not needed. Therefore, the method is very user-friendly, especially for new sub-scribers.
In the same way, the voice reference created takes into account possible user speech variations between two sessions. For example, it happens often that the user's voice is less sure and quick during the first session than during subsequent ones, when the user feels more confident.
It is possible to use a very high number of utterances for the voice reference without making the procedure too uncomfortable for the user. Even if the voice reference is already sufficiently good for a reliable verification of the user's identity, it is still possible to take new utterances into account in order to improve even more the quality of the voice reference and to adapt it to the user's voice variations.


REFERENCES:
patent: 4827518 (1989-05-01), Feustel et al.
patent: 5274695 (1993-12-01), Green
patent: 5297194 (1994-03-01), Hunt et al.
patent: 5430827 (1995-07-01), Rissanen
patent: 5548647 (1996-08-01), Naik et al.
patent: 55486

Affiliated with

Also associated with

Rating

Speaker verification method does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Speaker verification method, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Speaker verification method will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-3036312