METHODS AND SYSTEMS FOR ADAPTING GRAMMARS IN HYBRID SPEECH RECOGNITION ENGINES FOR ENHANCING LOCAL SR PERFORMANCE

• US 20130132086A1
• Filed: 11/21/2012
• Published: 05/23/2013
• Est. Priority Date: 11/21/2011
• Status: Active Grant

• Current Assignee: Robert Bosch GmbH
• Original Assignee: Robert Bosch GmbH
• Inventors: Xu, Kui, Weng, Fuliang, Feng, Zhe
• Granted Patent: US 9,153,229 B2
• Time in Patent Office: Days
• US Class Current: 704/257
• CPC Class Codes: G10L 15/005 Language recognition
• CPC Class Codes: G10L 15/01 Assessment or evaluation of speech recognition systems
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 15/30 Distributed recognition, e.g. in client-server systems, for mobile phones or network applications

Abstract

A speech recognition method includes providing a processor communicatively coupled to each of a local speech recognition engine and a server-based speech recognition engine. A first speech input is inputted into the server-based speech recognition engine. A first recognition result from the server-based speech recognition engine is received at the processor. The first recognition result is based on the first speech input. The first recognition result is stored in a memory device in association with the first speech input. A second speech input is inputted into the local speech recognition engine. The first recognition result is retrieved from the memory device. A second recognition result is produced by the local speech recognition engine. The second recognition result is based on the second speech input and is dependent upon the retrieved first recognition result.

Claims

1. A speech recognition method, comprising the steps of:

providing a processor communicatively coupled to each of a local speech recognition engine and a server-based speech recognition engine;

inputting a first speech input into the server-based speech recognition engine;

receiving at the processor a first recognition result from the server-based speech recognition engine, the first recognition result being based on the first speech input;

storing the first recognition result in a memory device, the first recognition result being stored in association with the first speech input;

inputting a second speech input into the local speech recognition engine;

retrieving the first recognition result from the memory device; and

producing a second recognition result by the local speech recognition engine, the second recognition result being based on the second speech input and being dependent upon the retrieved first recognition result.

2. The method of claim 1, comprising the further step of receiving at the processor a confidence score from the server-based speech recognition engine, the confidence score being indicative of a degree of confidence of the server-based speech recognition engine in a correctness of the first recognition result, the storing step being dependent upon the confidence score.

3. The method of claim 2, comprising the further step of counting a number of times that the confidence score exceeds a threshold level, the second recognition result being dependent upon the retrieved first recognition result only if the number of times that the confidence score exceeds the threshold level is greater than a predetermined number.

4. The method of claim 1, comprising the further step of receiving at the processor a confirmation from a user of a correctness of the first recognition result, the storing step being dependent upon the confirmation.

5. The method of claim 4, comprising the further step of counting a number of times that the user confirms the correctness of the first recognition result, the second recognition result being dependent upon the retrieved first recognition result only if the number of times that the user has confirmed the correctness of the first recognition result is greater than a predetermined number.

6. The method of claim 1, comprising the further step of determining whether there is a connection between the processor and a server hosting the server-based speech recognition engine, the producing step being dependent upon an existence of the connection.

7. The method of claim 1, wherein the producing step includes incorporating the first speech input into a grammar used in the server-based speech recognition engine.

8. The method of claim 7, wherein the grammar is dependent upon a domain of the first speech input.

9. The method of claim 7, wherein the incorporating step includes replacing a proper name in the first speech input with a corresponding name category to thereby create a modified first speech input.

10. The method of claim 9, wherein the incorporating step includes generating a shallow grammar that covers the modified first speech input.

11. The method of claim 1, comprising the further steps of:

comparing the first recognition result from the server-based speech recognition engine to the second recognition result from the local speech recognition engine;

determining whether the first recognition result matches the second recognition result;

inputting a third speech input into the local speech recognition engine; and

refraining from inputting the third speech input into the server-based speech recognition engine only if the first recognition result matches the second recognition result.

12. A speech recognition arrangement comprising:

a local speech recognition engine;

a server including a server-based speech recognition engine;

a user interface including a microphone;

a processor communicatively coupled to each of the local speech recognition engine, the server-based speech recognition engine, and the user interface, the user interface being configured to capture a first speech input by use of the microphone and transmit the first speech input to the processor, the processor being configured to input the first speech input into the server-based speech recognition engine, the server-based speech recognition engine being configured to output a first recognition result based on the first speech input; and

a memory device configured to store the first recognition result in a memory device in association with the first speech input, wherein the local speech recognition engine is configured to receive a second speech input and produce a second recognition result based on the second speech input and dependent upon the first recognition result.

13. The arrangement of claim 12 wherein the processor is configured to determine whether there is a connection between the processor and the server, the local speech recognition engine being configured to produce the second recognition result dependent upon an existence of the connection.

14. A speech recognition method, comprising the steps of:

using a microphone to capture a plurality of first speech inputs;

inputting the first speech inputs into a server-based speech recognition engine;

receiving first recognition results from the server-based speech recognition engine, the first recognition results being based on the first speech inputs;

receiving a plurality of confidence scores from the server-based speech recognition engine, each of the confidence scores being indicative of a degree of confidence of the server-based speech recognition engine in a correctness of a respective said first recognition result;

presenting the first recognition results to a user;

receiving a plurality of confirmations from the user, each of the confirmations being indicative of whether a respective one of the first recognition results is correct;

storing the first recognition results in a memory device, each of the first recognition results being stored in association with a respective one of the first speech inputs, the storing being dependent upon both the confidence scores and the confirmations from the user;

counting a number of times that both the confidence score exceeds a threshold level and the user confirms the correctness of the first recognition result;

inputting a second speech input into the local speech recognition engine;

retrieving at least one of the first recognition results from the memory device; and

producing a second recognition result by the local speech recognition engine, the second recognition result being;

based on the second speech input; and

dependent upon the retrieved at least one first recognition result only if the number of times that both the confidence score exceeds the threshold level and the user has confirmed the correctness of the first recognition result is greater than a predetermined number.

15. The method of claim 14, comprising the further step of determining whether there is a connection between the processor and a server hosting the server-based speech recognition engine, the local speech recognition engine producing the second recognition result if the connection does not exist, and the server-based speech recognition engine producing the second recognition result if the connection does exist.

16. The method of claim 14, wherein the producing step includes incorporating the first speech input into a grammar used in the server-based speech recognition engine.

17. The method of claim 16, wherein the grammar is dependent upon a domain of the first speech input.

18. The method of claim 16, wherein the incorporating step includes replacing a proper name in the first speech input with a corresponding name category to thereby create a modified first speech input.

19. The method of claim 18, wherein the incorporating step includes generating a shallow grammar that covers the modified first speech input.

20. The method of claim 14, comprising the further steps of:

comparing at least one of the first recognition results from the server-based speech recognition engine to the second recognition result from the local speech recognition engine;

determining whether the at least one first recognition result matches the second recognition result;

inputting a third speech input into the local speech recognition engine; and

refraining from inputting the third speech input into the server-based speech recognition engine only if the at least one first recognition result matches the second recognition result.