Load-adjusted speech recognition

• US 6,862,570 B2
• Filed: 04/28/2003
• Issued: 03/01/2005
• Est. Priority Date: 06/09/2000
• Status: Expired due to Term

• Current Assignee: ScanSoft, Inc. n/k/a Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: ScanSoft, Inc. n/k/a Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Schalkwyk, Johan
• Primary Examiner(s): McFadden, Susan
• Assistant Examiner(s): Opsasnick, Michael N.
• Application Number:
  US10/424,617
• Publication Number:
  US 20030195748A1
• Time in Patent Office: 673 Days
• Field of Search: 704/231, 704/270, 704/277
• US Class Current: 704/270
• CPC Class Codes: G10L 15/285 Memory allocation or algorithm optimisation to reduce hardware requirements

Abstract

A speech recognition system includes a user interface configured to provide signals indicative of a user'"'"'s speech. A speech recognizer of the system includes a processor configured to use the signals from the user interface to perform speech recognition operations to attempt to recognize speech indicated by the signals. A control mechanism is coupled to the voice recognizer and is configured to affect processor usage for speech recognition operations in accordance with a loading of the processor.

Claims

1. A speech recognition system comprising:

user interface logic configured to provide speech signals indicative of user speech;

speech recognition logic, responsive to processing parameters, wherein the processing parameters have an associated induced computational loading of a processor and an associated accuracy of speech recognition;

computational load-adjusting logic, responsive to a processor load variable representative of the computational loading of the processor, to adjust said processing parameters and thereby adjust computational loading of the processor and the accuracy of speech recognition

wherein the processor load variable identifies one of a plurality of processing classes including idle, normal, busy, and pegged classes, each with an associated range of computational loading of the processor.

2. The system of claim 1, wherein the speech recognition system includes scheduling logic to invoke the computational load-adjusting logic before the speech recognition logic.

3. The system of claim 1, wherein speech signals are comprised of multiple utterances and the computational load-adjusting logic is configured to determine the processor load variable at the beginning of each utterance.

4. The system of claim 1, wherein the processing parameters include at least one of a fast-match threshold, an across-word pruning value, and a short-list depth value.

5. The system of claim 1, wherein the processing parameters adjust the computational loading of the processor for recognizing speech to an idle limit, a normal limit, a busy limit, and a pegged limit when the processing class is identified to be idle, normal, busy, and pegged, respectively.

6. The system of claim 5, wherein the idle limit, normal limit, and pegged limit are adjusted in relation to the busy limit.

7. The system of claim 1, wherein the speech recognition logic includes logic to generate one or more acoustic models and the computational load-adjusting logic includes logic to identify via the processing parameters the acoustic model to be used by the speech recognition logic.

8. The system of claim 7, wherein the processing parameters identify a modeling technique to be used by the speech recognition logic to generate one or more acoustic models.

9. The system of claim 8, wherein each modeling technique has an associated computational loading of the processor and an associated accuracy of speech recognition.

10. A method of adaptive speech recognition, the method comprising:

receiving signals indicative of speech;

receiving a processor load variable representative of computational loading of a processor that identifies one of a plurality of processing classes of computational loading, the processor load variable further identifying one of a plurality of processing classes including idle, normal, busy, and pegged classes, each with an associated range of computational loading of the;

adjusting processing parameters in response to said processor load variable in order to adjust computational loading of the processor and accuracy of speech recognition;

recognizing speech signals according to the processing parameters.

11. The method of claim 10, wherein the processing parameters control the tradeoff between the computational loading of the processor and the accuracy of speech recognition.

12. The method of claim 10, wherein receiving signals indicative of speech includes receiving multiple utterances and receiving of the processor load variable occurs at a beginning of each utterance.

13. The method of claim 10, wherein the adjusting of the processing parameters occurs before the recognizing of the speech signals.

14. The method of claim 10, wherein adjusting the processing parameters further comprises adjusting the computational loading of the processor for recognizing speech to an idle limit, a normal limit, a busy limit, and a pegged limit when the processing class is identified to be idle, normal, busy, and pegged, respectively, and wherein the idle limit, normal limit, and pegged limit are adjusted in relation to the busy limit.

15. The method of claim 10, wherein the processing parameters include at last one of a fast-match threshold, an across-word pruning value, and a short-list depth value.

16. A computer-readable medium carrying one or more sequences of instructions for causing a computer to carry out the steps of:

receiving signals indicative of speech;

identifying one of a plurality of processing classes, idle, normal, busy, and pegged, each with an associated range of computational loading of the processor

receiving a processor load variable representative of computational loading of a processor that identifies one of the plurality of processing classes of computational loading;

adjusting processing parameters in response to said processor load variable in order to adjust computational loading of the processor and accuracy of speech recognition;

recognizing speech signals according to the processing parameters.

17. The computer-readable medium of claim 16, further including instructions for causing a computer to carry out the steps of:

receiving signals indicative of speech which include multiple utterances and adjusting the processing parameters at the beginning of each utterance.

18. The computer-readable medium of claim 16, further including instructions for causing a computer to carry out the steps of:

adjusting computational loading of the processor for recognizing speech to an idle limit, a normal limit, a busy limit, and a pegged limit when the processor class is identified to be idle, normal, busy, and pegged, respectively.

19. The computer-readable medium of claim 18, further including instructions for causing a computer to carry out the steps of:

adjusting the idle limit, normal limit, and pegged limit in relation to the busy limit.

20. The computer-readable medium of claim 16, wherein the processing parameters include at least one of a fast-match threshold, an across-word pruning value, and a short-list depth value.