Policy analysis framework for conversational biometrics
First Claim
1. A method of assessing error rates and tuning parameters associated with policy based procedures for verification of a plurality of known speakers through use of conversational biometrics, evolution of the policy based procedures being controlled by a state machine and each of the plurality of known speakers has an associated acoustic component and a knowledge profile, the method comprising the steps of:
- performing a policy based verification using a finite state machine defined by a set of states, each state having a specified set of questions and a plurality of possible state transitions, a state transition being performed if a condition of a state transition path associated with the state transition is satisfied;
determining allowability of the condition;
generating a state transition path through the state machine;
identifying all allowable state transition paths;
developing a probabilistic analysis of the behavior of the state machine, wherein each allowable state transition path is assigned a probability value;
generating a Receiver Operating Map (ROM) by using said probability values;
mapping a set of thresholds/parameters to a false accept rate and a false reject rate by using a multidimensional acoustic Receiver Operating Curve and probabilities of knowledge error, wherein the ROM is a function of a plurality of the thresholds/parameters with an output selected from one of accept rate and reject rate; and
tuning said thresholds/parameters by visually or analytically examining and optimizing over the ROM.
3 Assignments
0 Petitions
Accused Products
Abstract
The present invention describes a framework for the analytical and visual analysis and tuning of complex speaker verification systems guided by a policy finite state machine. The Receiver Operating Curve associated with the acoustic speaker recognition task is transformed into a multi-dimensional Receiver Operating Map (ROM), which results from a probabilistic analysis of the policy state machine. A Detection Cost Function (DCF) Map can be similarly generated. Results indicating that optimization over this surface (or the ROM) is an appropriate way to set thresholds are given.
45 Citations
20 Claims
-
1. A method of assessing error rates and tuning parameters associated with policy based procedures for verification of a plurality of known speakers through use of conversational biometrics, evolution of the policy based procedures being controlled by a state machine and each of the plurality of known speakers has an associated acoustic component and a knowledge profile, the method comprising the steps of:
-
performing a policy based verification using a finite state machine defined by a set of states, each state having a specified set of questions and a plurality of possible state transitions, a state transition being performed if a condition of a state transition path associated with the state transition is satisfied; determining allowability of the condition; generating a state transition path through the state machine; identifying all allowable state transition paths; developing a probabilistic analysis of the behavior of the state machine, wherein each allowable state transition path is assigned a probability value; generating a Receiver Operating Map (ROM) by using said probability values; mapping a set of thresholds/parameters to a false accept rate and a false reject rate by using a multidimensional acoustic Receiver Operating Curve and probabilities of knowledge error, wherein the ROM is a function of a plurality of the thresholds/parameters with an output selected from one of accept rate and reject rate; and tuning said thresholds/parameters by visually or analytically examining and optimizing over the ROM. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16)
-
-
17. A method of assessing error rates and tuning parameters associated with policy based procedures for verification of a plurality of known speakers through use of conversational biometrics, evolution of the policy based procedures being controlled by a state machine and each of the plurality of known speakers has an associated acoustic component and a knowledge profile, the method comprising the steps of:
-
performing a policy based verification using a finite state machine defined by a set of states, each state having a specified set of questions and a plurality of possible state transitions, a state transition being performed if a condition of a state transition path associated with the state transition is satisfied; generating a Receiver Operating Map (ROM) by using a plurality of probability values; mapping a set of thresholds/parameters to a false accept rate and a false reject rate by using a multidimensional acoustic Receiver Operating Curve and probabilities of knowledge error, wherein the ROM is a function of a plurality of the thresholds/parameters with an output selected from one of accept rate and reject rate; and utilizing the mapped set of thresholds/parameters in a speaker recognition process. - View Dependent Claims (18, 19, 20)
-
Specification