Speaker identification employing a confidence measure that uses statistical properties of N-best lists
View Patent ↗Methods and arrangements for facilitating speaker identification. At least one N-best list is generated based on input speech, a system output is posited based on the input speech, and a determination is made, via at least one property of the N-best list, as to whether the posited system output is inconclusive.
1. An apparatus for facilitating speaker identification, said apparatus comprising:
an arrangement for accepting input speech;
an arrangement for generating at least one N-best list based on the input speech;
an arrangement for positing a system output based on the input speech; and
an arrangement for ascertaining, via at least one statistical property of the N-best list, whether the posited system output is inconclusive,
wherein said ascertaining arrangement is adapted to generate at least two statistical models of the at least one N-best list in ascertaining whether the posited system output is inconclusive.
2. The apparatus according to claim 1 , wherein said ascertaining arrangement is adapted to combine results from the at least two statistical models to yield a confidence measure in ascertaining whether the posited system output is inconclusive.
3. The apparatus according to claim 2 , wherein said ascertaining arrangement is adapted to compare the confidence measure to a threshold value to determine whether the posited system output is inconclusive.
4. The apparatus according to claim 3 , wherein said ascertaining arrangement is further adapted to prompt the collection of additional input speech from the same speaker and reevaluate whether a new posited system output relating to the additional input speech is inconclusive.
5. The apparatus according to claim 1 , wherein one of said at least two statistical models relates to scores in the at least one N-best list.
6. The apparatus according to claim 5 , wherein said one of said at least two statistical models involves the calculation of a ratio of Gaussian densities relating to scores in the at least one N-best list.
7. The apparatus according to claim 1 , wherein one of said at least two statistical models relates to identities in the at least one N-best list.
8. The apparatus according to claim 7 , wherein said one of said at least two statistical models involves the estimation of the likelihood of identities in the at least one N-best list.
9. A method of facilitating speaker identification, said method comprising the steps of:
accepting input speech;
generating at least one N-best list based on the input speech;
positing a system output based on the input speech; and
ascertaining, via at least one statistical property of the N-best list, whether the posited system output is inconclusive,
wherein said ascertaining step comprises generating at least two statistical models of the at least one N-best list in ascertaining whether the posited system output is inconclusive.
10. The method according to claim 9 , wherein said ascertaining step comprises combining results from the at least two statistical models to yield a confidence measure in ascertaining whether the posited system output is inconclusive.
11. The method according to claim 10 , wherein said ascertaining step comprises comparing the confidence measure to a threshold value to determine whether the posited system output is inconclusive.
12. The method according to claim 11 , wherein said ascertaining step comprises prompting the collection of additional input speech from the same speaker and reevaluating whether a new posited system output relating to the additional input speech is inconclusive.
13. The method according to claim 9 , wherein one of said at least two statistical models relates to scores in the at least one N-best list.
14. The method according to claim 13 , wherein said one of said at least two statistical models involves the calculation of a ratio of Gaussian densities relating to scores in the at least one N-best list.
15. The method according to claim 9 , wherein one of said at least two statistical models relates to identities in the at least one N-best list.
16. The method according to claim 15 , wherein said one of said at least two statistical models involves the estimation of the likelihood of identities in the at least one N-best list.
17. A program storage device readable by machine, tangibly embodying a program of instructions executable by the machine to perform method steps for facilitating speaker identification, said method comprising the steps of:
accepting input speech;
generating at least one N-best list based on the input speech;
positing a system output based on the input speech; and
ascertaining, via at least one statistical property of the N-best list, whether the posited system output is inconclusive,
wherein said ascertaining step comprises generating at least two statistical models of the at least one N-best list in ascertaining whether the posited system output is inconclusive.