Method and Apparatus Including Parallell Processes for Voice Recognition

• US 20140278416A1
• Filed: 07/31/2013
• Published: 09/18/2014
• Est. Priority Date: 03/12/2013
• Status: Active Grant

• Current Assignee: Google Technology Holdings LLC (Alphabet Inc.)
• Original Assignee: Motorola Mobility LLC (Lenovo Group Ltd.)
• Inventors: Schuster, Adrian M., Clark, Joel A., Davis, Giles T., Ivanov, Plamen A., Zurek, Robert A.
• Granted Patent: US 9,542,947 B2
• Time in Patent Office: Days
• US Class Current: 704/246
• CPC Class Codes: G10L 15/32 Multiple recognisers used in sequence or in parallel; Score combination systems therefor, e.g. voting systems
• CPC Class Codes: G10L 17/00 Speaker identification or verification techniques

Abstract

A method and apparatus for voice recognition performed in a voice recognition block comprising a plurality of voice recognition stages. The method includes receiving a first plurality of voice inputs, corresponding to a first phrase, into a first voice recognition stage of the plurality of voice recognition stages, wherein multiple ones of the voice recognition stages includes a plurality of voice recognition modules and multiples ones of the voice recognition stages perform a different type of voice recognition processing, wherein the first voice recognition stage processes the first plurality of voice inputs to generate a first plurality of outputs for receipt by a subsequent voice recognition stage. The method further includes, receiving by each subsequent voice recognition stage a plurality of outputs from a preceding voice recognition stage, wherein a plurality of final outputs is generated by a final voice recognition stage from which to approximate the first phrase.

Claims

1. A method for voice recognition performed in a voice recognition block comprising a plurality of voice recognition stages, the method comprising:

receiving a first plurality of voice inputs, corresponding to a first phrase, into a first voice recognition stage of the plurality of voice recognition stages, wherein multiple ones of the voice recognition stages comprise a plurality of voice recognition modules and multiples ones of the voice recognition stages perform a different type of voice recognition processing, wherein the first voice recognition stage processes the first plurality of voice inputs to generate a first plurality of outputs for receipt by a subsequent voice recognition stage of the plurality of voice recognition stages;

receiving by each subsequent voice recognition stage a plurality of outputs from an immediately preceding voice recognition stage, wherein each subsequent voice recognition stage processes a plurality of outputs from the immediately preceding voice recognition stage to generate a plurality of corresponding subsequent outputs, wherein a plurality of final outputs each containing a corresponding phrase is generated by a final voice recognition stage from which to approximate the first phrase corresponding to the first plurality of voice inputs.

2. The method of claim 1 further comprising selecting at least one of the final outputs as approximating the first phrase.

3. The method of claim 2, wherein the at least one selected final output has a highest likelihood of containing the first phrase corresponding to the plurality of voice inputs.

4. The method of claim 3, wherein the highest likelihood is determined based on a highest confidence score calculated for the at least one selected final output.

5. The method of claim 3, wherein the highest likelihood is determined based on a highest aggregate confidence score calculated for the at least one selected final output.

6. The method of claim 3, wherein the highest likelihood is determined based on the at least one selected final output containing a most frequently appearing phrase across the plurality of final outputs.

7. The method of claim 1, wherein the plurality of final outputs is generated by a plurality of different paths through the voice recognition block, wherein each path comprises processing performed by one of the voice recognition modules from each voice recognition stage of the voice recognition block.

8. The method of claim 7 further comprising selecting a subset of the plurality of different paths through the voice recognition block for processing a second plurality of voice inputs subsequently received into the first voice recognition stage.

9. The method of claim 8, wherein selecting the subset of the plurality of different paths is based at least in part on historical data determined while processing at least one plurality of inputs.

10. The method of claim 9, wherein the historical data comprises data indicating a name uttered by a user, a user call history, user command history, a user trigger, a user command or a call time of day.

11. The method of claim 9, wherein the historical data comprises data indicating a characteristic of a user that supplied the first phrase, wherein the characteristic indicates at least one of a language, dialect, or accent spoken by the user.

12. The method of claim 9, wherein the historical data comprises data indicating an environment within which a device containing the voice recognition block is operated.

13. The method of claim 1 further comprising performing signal conditioning processing, noise suppression processing, voice recognition engine processing, and voice recognition language processing in different ones of the plurality of voice recognition stages, to generate the plurality of final outputs.

14. The method of claim 13, wherein the voice recognition block processes the first plurality of voice inputs using at least one of:

a plurality of different signal conditioning modules that perform the signal conditioning processing;

a plurality of different noise suppression modules that perform the noise suppression processing;

a plurality of different voice recognition engine modules that perform the voice recognition engine processing;

or a plurality of different voice recognition language modules that perform the voice recognition language processing.

15. The method of claim 14 further comprising:

determining an adjusted paths through one or more of a set of voice recognition processing paths associated with at least one of the final outputs selected as approximating the first phrase through the voice recognition block by selecting an adjusted number of paths comprising processing using at least one of;

a subset of the plurality of different signal conditioning modules, a subset of the plurality of different noise suppression modules, a subset of the plurality of different voice recognition engine modules, or a subset of the plurality of different voice recognition language modules;

subsequently receiving into the first voice recognition stage a second plurality of voice inputs corresponding to a second phrase;

processing the second plurality of voice inputs in the voice recognition block using the reduced set of voice recognition processing paths.

16. The method of claim 1 further comprising selecting the plurality of inputs from a larger set of inputs to generate the plurality of final outputs.

17. A system for performing voice recognition, the system comprising:

a first voice recognition stage configured to receive a first plurality of voice channels corresponding to a first phrase and configured with at least one voice recognition module to process the first plurality of voice channels and generate a first plurality of outputs;

a set of subsequent voice recognition stages each configured with at least one voice recognition module to process a plurality of outputs from the immediately preceding voice recognition stage and to generate a plurality of corresponding subsequent outputs, a final voice recognition stage of the set of subsequent voice recognition stages that is configured to generate a plurality of final outputs, wherein multiple ones of the voice recognition stages comprise a plurality of voice recognition modules and multiples ones of the voice recognition stages are configured to perform a different type of voice recognition processing;

a selector module configured to select at least one of the final outputs to approximate the first phrase corresponding to the first plurality of voice channels.

18. The system of claim 17, wherein the selected at least one final output is selected based on one of confidence score calculated for the at least one selected final output, a highest aggregate confidence score calculated for the at least one selected final output, or a most frequently appearing phrase across the at least one final output.

19. The system of claim 17, further comprising a control module configured to select a subset of a plurality of different paths based on historical data comprising data indicating at least one of a characteristic of the user, whether a trigger has been spoken, whether a local command is uttered after the trigger, whether a network based command or search is required, an environment in which a device containing the voice recognition block is operating, a user call history, or a call time of day, wherein the selection of the subset of the plurality of different paths is performed in parallel or serially.

20. The system of claim 17, wherein the system comprises at least one voice recognition stage configured to perform signal conditioning processing, at least one voice recognition stage configured to perform noise suppression processing, and at least one voice recognition stage configured to perform voice recognition engine processing.

21. The system of claim 17, wherein the voice recognition block is configured to processes the first plurality of voice inputs using at least one of:

a plurality of different signal conditioning modules that perform the signal conditioning processing;

a plurality of different noise suppression modules that perform the noise suppression processing;

or a plurality of different voice recognition engine modules that perform the voice recognition engine processing.

22. The system of claim 17 further comprising at multiple instances of at least one of the modules in at least one of the voice recognition stages.