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.
-
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