Methods and systems for splitting a digital signal

Patent metadata

Field	Value
Publication number	US-9646613-B2
Application number	US-201314093200-A
Country	US
Kind code	B2
Filing date	Nov 29, 2013
Priority date	Nov 29, 2013
Publication date	May 9, 2017
Grant date	May 9, 2017

Abstract

A method for splitting a digital signal using prosodic features included in the signal is provided that includes calculating onset value locations in the signal. The onset values correspond to stress accents in the signal. Moreover, the method includes splitting, using a processor, the signal into a prosodic unit candidate sequence by superimposing the stress accent locations on the signal, and processing the sequence to include only true prosodic units.

First claim

What is claimed is:

1. A method for authenticating users comprising:
calculating signal features for an audio digital signal representing a user utterance, the signal features including onset values, the onset values correspond to stress accents in the signal;
splitting, by a processor, the signal into a prosodic unit candidate sequence based on the stress accent locations;
modifying, by the processor, the prosodic unit candidate sequence based on the signal features, the modified prosodic unit sequence including only true prosodic units
calculating a first probability that the true prosodic units match a prosodic unit sequence used to create a hidden Markov model for the user;
calculating a second probability that the true prosodic units match prosodic unit sequences used to create a Universal Background Model; and
successfully authenticating the user when the difference between the first probability and the second probability is greater than a decision threshold.

2. A method in accordance with claim 1 , said modifying step comprising:
calculating the energy level for each prosodic unit candidate in the prosodic unit candidate sequence; and
removing from the prosodic unit candidate sequence prosodic unit candidates having a low energy level.

3. A method in accordance with claim 1 , said modifying step comprising:
identifying borders, shared by prosodic unit candidates, located in a high level region of the signal; and
moving the identified borders to a low level energy region of the signal.

4. A method in accordance with claim 1 , further comprising:
creating a hidden Markov model in which each true prosodic unit defines a state in the model; and
optimizing the model.

5. A computing system for authenticating users comprising:
a processor; and
a memory configured to store data, said computing device being associated with a network and said memory being in communication with said processor and having instructions stored therein which, when read and executed by said processor cause said computing device to:
calculate signal features for an audio digital signal representing a user utterance, the signal features including onset values, the onset values correspond to stress accents in the signal;
split the signal into a prosodic unit candidate sequence based on the stress accent locations;
modify the prosodic unit candidate sequence based on the signal features, the modified prosodic unit sequence including only true prosodic units;
calculate a first probability that the true prosodic units match a prosodic unit sequence used to create a hidden Markov model for the user;
calculate a second probability that the true prosodic units match prosodic unit sequences used to create a Universal Background Model; and
successfully authenticate the user when the difference between the first probability and the second probability is greater than a decision threshold.

6. A computing system in accordance with claim 5 , wherein the instructions when executed by said processor further cause said computing device to:
calculate the energy level for each prosodic unit candidate in the prosodic unit candidate sequence; and
remove from the prosodic unit candidate sequence prosodic unit candidates having a low energy level.

7. A computing system in accordance with claim 5 , wherein the instructions when executed by said processor further cause said computing device to:
identify borders, shared by prosodic unit candidates, located in a high level region of the signal; and
move the identified borders to a low level energy region of the signal.

8. A computing system in accordance with claim 5 , wherein the instructions when executed by said processor further cause said computing device to:
create a hidden Markov model in which each true prosodic unit defines a state in the model; and
optimize the model.

9. A computing system in accordance with claim 5 , wherein the instructions when executed by said processor further cause said computing device to compute a decision score using the hidden Markov model.

10. A computing system in accordance with claim 5 , said computing device being a smart phone or a tablet computer.

11. A computing system in accordance with claim 5 , said computing device being an authentication computer system.

12. A computer program recorded on a non-transitory computer-readable recording medium included in a computing device for generating trustworthy authentication transaction results, the computer program being comprised of instructions, which when read and executed by the computing device, cause the computing device to:
calculate signal features for an audio digital signal representing a user utterance, the signal features including onset values, the onset values correspond to stress accents in the signal;
split the signal into a prosodic unit candidate sequence based on the stress accent locations;
modify the prosodic unit candidate sequence based on the signal features, the modified prosodic unit sequence including only true prosodic units;
calculate a first probability that the true prosodic units match a prosodic unit sequence used to create a hidden Markov model for the user;
calculate a second probability that the true prosodic units match prosodic unit sequences used to create a Universal Background Model; and
successfully authenticate the user when the difference between the first probability and the second probability is greater than a decision threshold.

13. A computer program in accordance with claim 12 wherein the instructions which when read and executed by the computing device, further cause the computing device to:
calculate the energy level for each prosodic unit candidate in the prosodic unit candidate sequence; and
remove from the prosodic unit candidate sequence prosodic unit candidates having a low energy level.

14. A computer program in accordance with claim 12 wherein the instructions which when read and executed by the computing device, further cause the computing device to:
identify borders, shared by prosodic unit candidates, located in a high level region of the signal; and
move the identified borders to a low level energy region of the signal.

15. A computer program in accordance with claim 12 wherein the instructions which when read and executed by the computing device, further cause the computing device to:
create a hidden Markov model in which each true prosodic unit defines a state in the model; and
optimize the model.

16. A computer program in accordance with claim 12 wherein the instructions which when read and executed by the computing device, further cause the computing device to compute a decision score using the hidden Markov model.

Assignees

• Blouet Raphael
• Daon Holdings Ltd

Inventors

• Blouet Raphael

Classifications

• H04M3/385

  using speech signals · CPC title

• G10L15/1807

  using prosody or stress · CPC title

• G10L15/04 Primary

  Segmentation; Word boundary detection · CPC title

• G10L17/04

  Training, enrolment or model building · CPC title

• G10L17/22 Primary

  Interactive procedures; Man-machine interfaces · CPC title