Hybrid arbitration system

1. A method for selecting a speech recognition result on a computing device, the method comprising:

receiving a first speech recognition result determined by the computing device;

receiving a first plurality of features, at least some of the features being determined using the first speech recognition result; and

determining whether to select the first speech recognition result or to wait for a second speech recognition result determined by a cloud computing service based at least in part on the first speech recognition result and the first plurality of features, wherein the determining of whether to select the first speech recognition result or to wait for a second speech recognition result includes computing, based at least in part on the first speech recognition result and the first plurality of features, a first confidence value for the first speech recognition result and comparing the first confidence value to a first predetermined threshold.

2. The method of claim 1 further comprising, based on the determining, waiting for the second speech recognition result.

3. The method of claim 2 further comprising:

receiving the second speech recognition result;

receiving a second plurality of features, at least some of the features being determined using the second speech recognition result; and

determining whether to select the first speech recognition result or the second speech recognition result based at least in part on the first speech recognition result, the first plurality of features the second speech recognition result, and the second plurality of features.

4. The method of claim 3 wherein determining whether to select the first speech recognition result or the second speech recognition result includes:

computing a second confidence value for the second speech recognition result based at least in part on the second speech recognition result and the second plurality of features; and

comparing the first confidence value to the second confidence value.

5. The method of claim 4 wherein computing the first confidence value and the second confidence value includes processing the first speech recognition result, the first plurality of features, in a classifier.

6. The method of claim 5 wherein the classifier is implemented as a neural network.

7. The method of claim 4 further comprising selecting the first speech recognition result if the first confidence value exceeds the second confidence value.

8. The method of claim 3 wherein at least some of the second plurality of features characterize a quality of the second speech recognition result.

9. The method of claim 3 wherein the second plurality of features includes one or more of a natural language understanding dialogue context, a natural language understanding recognized domain name, a natural language understanding recognized context name, a natural language understanding recognized action name, one or more natural language understanding recognized entity names, and an NBest word sequence.

10. The method of claim 1 wherein at least some of the first plurality of features characterize a quality of the first speech recognition result.

11. The method of claim 1 wherein the first plurality of features includes one or more of a natural language understanding dialogue context, a natural language understanding recognized domain name, a natural language understanding recognized context name, a natural language understanding recognized action name, one or more natural language understanding recognized entity names, and an NBest word sequence.

12. The method of claim 1 further comprising selecting the first speech recognition result if the first confidence value exceeds the first predetermined threshold.

13. The method of claim 1 further comprising waiting for the second speech recognition result if the first confidence value does not exceed the first predetermined threshold.

14. The method of claim 1 wherein the computing includes processing the first speech recognition result and the first plurality of features in a classifier.

15. The method of claim 14 wherein the classifier is implemented as a neural network.

16. The method of claim 1 wherein the first speech recognition result and the first plurality of features are determined based at least in part on user data stored on the computing device.

17. The method of claim 1 further comprising performing an action based at least in part on the selected speech recognition result.

18. A system for selecting a speech recognition result on a computing device, the system comprising:

an input for receiving a first speech recognition result determined by the computing device;

an input for receiving a first plurality of features, at least some of the features being determined using the first speech recognition result; and

one or more processors for processing the first speech recognition result and the first plurality of features to determine whether to select the first speech recognition result or to wait for a second speech recognition result determined by a cloud computing service based at least in part on the first speech recognition result and the first plurality of features, wherein the determining of whether to select the first speech recognition result or to wait for a second speech recognition result includes computing, based at least in part on the first speech recognition result and the first plurality of features, a first confidence value for the first speech recognition result and comparing the first confidence value to a first predetermined threshold.

19. Software stored on a non-transitory, computer-readable medium, the software including instructions for causing one or more processors to:

receive a first speech recognition result determined by the computing device;

receive a first plurality of features, at least some of the features being determined using the first speech recognition result; and

determine whether to select the first speech recognition result or to wait for a second speech recognition result determined by a cloud computing service based at least in part on the first speech recognition result and the first plurality of features, wherein the determining of whether to select the first speech recognition result or to wait for a second speech recognition result includes using the one or more processors to compute, based at least in part on the first speech recognition result and the first plurality of features, a first confidence value for the first speech recognition result and compare the first confidence value to a first predetermined threshold.