Method and system for non-intrusive speaker verification using behavior models

#1. Field of the Invention

#The present invention relates to natural language understanding systems, and more particularly to a method and system for non-intrusively verifying a user based on behavior of the user.

#2. Description of the Related Art

#The invention will be described in detail in the following description of preferred embodiments with reference to the following figures wherein:

#FIG. 1 is a block/flow diagram of an illustrative system/method employing behavior verification in accordance with the present invention; and

#FIG. 2 is a block diagram of an illustrative behavior verifier in accordance with the present invention.

#The present invention provides a method and system for continuously verifying the identity of the user based on how the user interacts with a target system. This may be performed by comparing the user's current behavior with past behavior. No additional dedicated input (beyond what is used for the initial verification) from the user is needed, and therefore the system is non-intrusive. Furthermore, verification can be done continuously, and if enough evidence is available for rejecting the user in the middle of a session, then the user can be cut-off before more damage is done.

#In alternate embodiments, even the initial verification is not necessary, and all users may be given base level access (to non-confidential information, for example), and upon additional verification through the non-intrusive process, full access may be given.

#In the present invention, a new dimension to the speaker (or user) verification paradigm is provided, by introducing a new criterion: how the user behaves. For example, a user who usually greets with a “Howdy”, may be distinguished from an imposter who greets with a “Hello” or a “How are you”, or from an imposter who begins the conversation without any greetings. Similarly, an imposter who is trying to search through several confidential documents may be distinguished from a legitimate user who does not typically perform such searches. Although any one single interaction with the system may be insufficient to make a decision, information gathered after several user-system interactions may be sufficient to make a valid verification decision.