System and method for optimizing matched voice biometric passphrases

1. A method for optimizing matched voice biometric passphrases, performed on a computing device having a processor, memory, and one or more code sets stored in the memory and executing in the processor, the method comprising:

generating, by the processor, a plurality of k candidate passphrases;

calculating, by the processor, a phonetic content of each candidate passphrase of the plurality of k candidate passphrases;

capturing a text-independent voiceprint of a user;

calculating, by the processor, a user phonetic content of a text-independent voiceprint of the user;

rank-ordering each of the plurality of k candidate passphrases based on how similar the phonetic content of each of the k candidate passphrases is to the user phonetic content of the text-independent voiceprint of the user;

identifying, by the processor, a subset of j passphrases based on the rank-ordering from the plurality of k candidate passphrases, wherein each of the subset of j passphrases meets a match threshold in phonetic content with the voiceprint of the user;

selecting, by the processor, a first one of the subset of j passphrases; and

presenting, by the processor, the selected first one of the subset of j passphrases to the user for use in the user authentication.

2. The method as in claim 1 , wherein calculating, by the processor, the phonetic content of each candidate passphrase of the plurality of k candidate passphrases comprises:

comparing, by the processor, each candidate passphrase of the plurality of k candidate passphrases against a pronunciation lexicon;

based on the comparison, identifying, by the processor, the pronunciation of each phoneme in each candidate passphrase of the plurality of k candidate passphrases; and

generating, by the processor, a phoneme string representing each candidate passphrase of the plurality of k candidate passphrases.

3. The method as in claim 1 , wherein calculating, by the processor, the phonetic content of the text-independent voiceprint of a user comprises:

comparing, by the processor, the text-independent voiceprint against a pronunciation lexicon;

based on the pronunciation lexicon, identifying, by the processor, the pronunciation of each phoneme in the text-independent voiceprint; and

generating, by the processor, a phoneme string representing the text-independent voiceprint.

4. The method as in claim 1 , wherein calculating, by the processor, the phonetic content of the text-independent voiceprint of a user comprises:

extracting, by the processor, a Universal Background Model (UBM) state occupancy vector from the text-independent voiceprint; and

based on the UBM state occupancy, generating, by the processor, a phoneme string for the text-independent voiceprint.

5. The method as in claim 4 , further comprising:

transforming the phonetic content of each candidate passphrase of the plurality of k candidate passphrases into an estimated state occupancy vector;

comparing the distance between each estimated state occupancy vector and the Universal Background Model (UBM) state occupancy vector;

ranking, by the processor, each candidate passphrase of the plurality of k candidate passphrases based on the distance; and

selecting, by the processor, the subset of j passphrases within a threshold distance.

6. The method as in claim 1 , further comprising:

prompting, by the processor, the user to verbally articulate the selected first one of the subset of j passphrases to the user for use in the user authentication; and

authenticating, by the processor, the user based on the verbally articulated selected first one of the subset of j passphrases and the voiceprint of the user.

7. The method as in claim 1 , further comprising:

selecting, by the processor, a second one of the subset of j passphrases; and

presenting the selected second one of the subset of j passphrases to the user for use in a subsequent user authentication.

8. The method as in claim 7 , wherein at least one of the first one of the subset of j passphrases and the second one of the subset of j passphrases is selected randomly.

9. A system for optimizing matched voice biometric passphrases, comprising:

a computer having a processor and a memory; and

one or more code sets stored in the memory and executed by the processor, which configure the processor to:

generate a plurality of k candidate passphrases;

calculate a phonetic content of each candidate passphrase of the plurality of k candidate passphrases;

capture a text-independent voiceprint of a user;

calculate user phonetic content of a text-independent voiceprint of the user;

rank each of the k candidate passphrases based on how similar the phonetic content of each of the k candidate passphrases is to the user phonetic content of the text-independent voiceprint of the user;

identify a subset of j passphrases based on the rank-ordering from the plurality of k candidate passphrases, wherein each of the subset of j passphrases meets a match threshold in phonetic content with the voiceprint of the user;

select a first one of the subset of j passphrases; and

present the selected first one of the subset of j passphrases to the user for use in the user authentication.

10. The system as in claim 9 , wherein, when calculating the phonetic content of each candidate passphrase of the plurality of k candidate passphrases, the processor is configured to:

compare each candidate passphrase of the plurality of k candidate passphrases against a pronunciation lexicon;

based on the comparison, identify the pronunciation of each phoneme in each candidate passphrase of the plurality of k candidate passphrases; and

generate a phoneme string representing each candidate passphrase of the plurality of k candidate passphrases.

11. The system as in claim 9 , wherein, when calculating the phonetic content of the text-independent voiceprint of a user, the processor is configured to:

compare the text-independent voiceprint against a pronunciation lexicon;

based on the pronunciation lexicon, identify the pronunciation of each phoneme in the text-independent voiceprint; and

generate a phoneme string representing the text-independent voiceprint.

12. The system as in claim 9 , wherein, when calculating the phonetic content of the text-independent voiceprint of a user, the processor is configured to:

extract a Universal Background Model (UBM) state occupancy vector from the text-independent voiceprint; and

based on the UBM state occupancy, generate a phoneme string for the text-independent voiceprint.

13. The system as in claim 12 , the processor further configured to:

transform the phonetic content of each candidate passphrase of the plurality of k candidate passphrases into an estimated state occupancy vector;

compare the distance between each estimated state occupancy vector and the Universal Background Model (UBM) state occupancy vector;

rank each candidate passphrase of the plurality of k candidate passphrases based on the distance; and

select the subset of j passphrases within a threshold distance.

14. The system as in claim 9 , the processor further configured to:

prompt the user to verbally articulate the selected first one of the subset of j passphrases to the user for use in the user authentication; and

authenticate the user based on the verbally articulated selected first one of the subset of j passphrases and the voiceprint of the user.

15. The system as in claim 9 , further configured to:

select a second one of the subset of j passphrases; and

present the selected second one of the subset of j passphrases to the user for use in a subsequent user authentication.

16. The method as in claim 15 , wherein at least one of the first one of the subset of j passphrases and the second one of the subset of j passphrases is selected randomly.

17. A method for creating voice passphrases, the method comprising:

generating, by a computer processor, a plurality of candidate phrases, each candidate phrase comprising a plurality of words;

calculating, by the processor, a phoneme sequence of each phrase;

capturing a text-independent voiceprint of a user;

calculating, by the processor, a user phoneme sequence of the text-independent voiceprint of the user;

rank-ordering each of the plurality of candidate phrases based on how similar the phoneme sequence of each of the candidate phrases is to the user phoneme sequence;

identifying, by the processor, a subset of phrases from the plurality of phrases based on the rank-ordering, wherein each of the subset of phrases matches in phonetic content based on a threshold with the text-independent voiceprint of the user;

selecting, by the processor, a first one of the subset of phrases;

providing, by the processor, the selected phrase to the user; and

receiving authentication from the user voicing the selected phrase.

18. The method as in claim 17 , wherein identifying, by the processor, the subset of phrases from the plurality of phrases comprises:

ranking, by the processor, the plurality of phrases by similarity in phonetic content with the voiceprint of the user; and

selecting, by the processor, the subset of phrases which meet a match threshold in phonetic content with the voiceprint of the user.