Context-aware unit selection

• US 20090132253A1
• Filed: 11/20/2007
• Published: 05/21/2009
• Est. Priority Date: 11/20/2007
• Status: Active Grant

• Current Assignee: Apple Inc.
• Original Assignee: Apple Inc.
• Inventors: Bellegarda, Jerome
• Granted Patent: US 8,620,662 B2
• Time in Patent Office: Days
• US Class Current: 704/258
• CPC Class Codes: G10L 13/06 Elementary speech units used in speech synthesisers; Concatenation rules

Abstract

Methods and apparatuses to perform context-aware unit selection for natural language processing are described. Streams of information associated with input units are received. The streams of information are analyzed in a context associated with first candidate units to determine a first set of weights of the streams of information. A first candidate unit is selected from the first candidate units based on the first set of weights of the streams of information. The streams of information are analyzed in the context associated with second candidate units to determine a second set of weights of the streams of information. A second candidate unit is selected from second candidate units to concatenate with the first candidate unit based on the second set of weights of the streams of information.

Claims

1. A machine-implemented method, comprising:

analyzing streams of information in a context associated with first candidate units to determine a distribution of the streams of information over the first candidate units;

determining a first set of weights of streams of information according to the distribution; and

selecting a first candidate unit from the first candidate units based on the first set of weights of the streams of information.

2. The machine-implemented method as in claim 1, further comprising:

analyzing the streams of information in the context associated with second candidate units to determine a second set of weights of streams of information; and

selecting a second candidate unit from second candidate units to concatenate with the first candidate unit based on the second set of weights of streams of information.

3. The machine-implemented method of claim 1, wherein the analyzing includes:

weighting a first stream of information of the streams of information higher if the first stream of information provides a high discrimination between the first candidate units.

4. The machine-implemented method of claim 1, wherein the analyzing includes:

weighting a second stream of information of the streams of information lower if the second stream of information provides a low discrimination between the first candidate units.

5. The machine-implemented method of claim 1, wherein a stream of information of the streams of information is to represent a characteristic associated with an input unit.

6. A machine-implemented method, comprising:

determining first scores associated with streams of information for first candidate units associated with a first input unit;

generating a first matrix of the first scores for the first candidate units;

determining a first set of weights using the first matrix;

determining first final costs for the first candidate units using the first set of weights;

selecting a first candidate unit from the first candidate units based on the first final costs.

7. The machine-implemented method of claim 6, further comprising:

normalizing the scores across the candidate units.

8. The machine-implemented method of claim 6, wherein the determining the first set of weights includes maximizing the first final costs for the first candidate units.

9. The machine-implemented method of claim 6, wherein the first candidate unit has a minimal final cost.

10. The machine-implemented method of claim 6, wherein a stream of information of the streams of information includes a cost function.

11. The machine-implemented method of claim 6, further comprising:

determining second scores associated with streams of information for second candidate units associated with a second input unit;

generating a second matrix of the second scores for second candidate units;

determining a first set of weights using the second matrix;

determining second final costs for second candidate units using the second set of weights;

selecting a second candidate unit from the second candidate units based on the second final costs.

12. A machine-readable medium containing executable program instructions which cause a data processing system to perform operations comprising:

analyzing streams of information in a context associated with first candidate units to determine a distribution of the streams of information over the first candidate units;

determining a first set of weights of streams of information; and

selecting a first candidate unit from the first candidate units based on the first set of weights of the streams of information.

13. The machine-readable medium of claim 12, further including data that cause the data processing system to perform operations comprising:

analyzing the streams of information in the context associated with second candidate units to determine a second set of weights of the streams of information; and

selecting a second candidate unit from second candidate units to concatenate with the first candidate unit based on the second set of weights of the streams of information.

14. The machine-readable medium of claim 12, wherein the analyzing includes weighting a first stream of information of the streams of information higher if the first stream of information provides a high discrimination between the first candidate units.

15. The machine-readable medium of claim 12, wherein the analyzing includes weighting a second stream of information of the streams of information lower if the second stream of information provides a low discrimination between the first candidate units.

16. The machine-readable medium of claim 12, wherein a stream of information of the streams of information is to represent a characteristic associated with an input unit.

17. A machine-readable medium containing executable program instructions which cause a data processing system to perform operations comprising:

determining first scores associated with streams of information for first candidate units associated with a first input unit;

generating a first matrix of the first scores for the first candidate units;

determining a first set of weights using the first matrix;

determining first final costs for the first candidate units using the first set of weights;

selecting a first candidate unit from the first candidate units based on the first final costs.

18. The machine-readable medium of claim 17, further including data that cause the data processing system to perform operations comprising:

normalizing the scores across the candidate units.

19. The machine-readable medium of claim 17, wherein the determining the first set of weights includes maximizing the first final costs for the first candidate units.

20. The machine-readable medium of claim 17, wherein the first candidate unit has a minimal final cost.

21. The machine-readable medium of claim 17, wherein a stream of information of the streams of information includes a cost function.

22. The machine-readable medium of claim 17, further including data that cause the data processing system to perform operations comprising:

determining second scores associated with streams of information for second candidate units associated with a second input unit;

generating a second matrix of the second scores for second candidate units;

determining a first set of weights using the second matrix;

determining second final costs for second candidate units using the second set of weights;

selecting a second candidate unit from the second candidate units based on the second final costs.

23. A data processing system, comprising:

means for analyzing streams of information in a context associated with first candidate units to determine a distribution of the streams of information;

means for determining a first set of weights of the streams of information according to the distribution; and

means for selecting a first candidate unit from the first candidate units based on the first set of weights of the streams of information.

24. The data processing system as in claim 23, further comprising:

means for analyzing the streams of information in the context associated with second candidate units to determine a second set of weights of the streams of information; and

means for selecting a second candidate unit from second candidate units to concatenate with the first candidate unit based on the second set of weights of the streams of information.

25. The data processing system of claim 23 further comprising:

means for determining first scores associated with the streams of information for the first candidate units associated with a first input unit;

means for generating a first matrix of the first scores for the first candidate units;

means for determining a first set of weights using the first matrix;

means for determining first final costs for the first candidate units using the first set of weights; and

means for selecting the first candidate unit from the first candidate units based on the first final costs.