DETECTING POTENTIAL SIGNIFICANT ERRORS IN SPEECH RECOGNITION RESULTS

• US 20150248882A1
• Filed: 05/15/2015
• Published: 09/03/2015
• Est. Priority Date: 07/09/2012
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Ganong, William F. III, Vemula, Raghu, Fleming, Robert
• Granted Patent: US 9,818,398 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G10L 15/01 Assessment or evaluation of speech recognition systems
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 15/24 Speech recognition using non-acoustical features
• CPC Class Codes: G10L 2015/085 Methods for reducing search complexity, pruning

Abstract

In some embodiments, recognition results produced by a speech processing system (which may include two or more recognition results, including a top recognition result and one or more alternative recognition results) based on an analysis of a speech input, are evaluated for indications of potential errors. In some embodiments, the indications of potential errors may include discrepancies between recognition results that are meaningful for a domain, such as medically-meaningful discrepancies. The evaluation of the recognition results may be carried out using any suitable criteria, including one or more criteria that differ from criteria used by an ASR system in determining the top recognition result and the alternative recognition results from the speech input. In some embodiments, a recognition result may additionally or alternatively be processed to determine whether the recognition result includes a word or phrase that is unlikely to appear in a domain to which speech input relates.

Claims

1. A method of processing results of a recognition by an automatic speech recognition (ASR) system on a speech input, the results comprising two or more results identified by the ASR system as likely to be an accurate recognition result for the speech input, the method comprising:

evaluating the two or more results using at least one criterion that differs from criteria used by the ASR system in determining the two or more results; and

in response to determining that the at least one criterion is met by the two or more results, triggering an alert concerning one of the two or more results.

2. The method of claim 1, wherein the evaluating the two or more results using the at least one criterion comprises evaluating the two or more results for medically-meaningful discrepancies between the two or more results.

3. The method of claim 1, wherein the evaluating and triggering are performed by an entity other than the ASR system.

4. The method of claim 1, wherein:

the two or more results comprise a first recognition result identified by the ASR system as most likely to be a correct recognition result for the speech input, and

the evaluating the two or more results using the at least one criterion comprises determining whether the two or more results comprise an indication of a potential error in the first recognition result that may cause a meaning of the first recognition result to differ from a meaning of the speech input.

5. The method of claim 4, wherein:

the two or more results comprise a first recognition result and a second recognition result;

the method further comprises semantically interpreting each of the first recognition result and the second recognition result to determine at least one first fact expressed in the first recognition result and at least one second fact expressed in the second recognition result; and

the determining whether the two or more results comprise an indication of a potential error that may cause the meaning of the first recognition result to differ from the meaning of the speech input comprises determining whether there is a difference between the at least one first fact and the at least one second fact.

6. The method of claim 4, wherein determining whether the two or more results comprise an indication of a potential error that may cause the meaning of the first recognition result to differ from a meaning of the speech input comprises determining whether the first recognition result includes a first member of a set of words or phrases, each member of the set comprising a word or phrase and being associated with at least one other member of the set, and determining whether the second recognition result includes at least one other member associated with the first member of the set.

7. The method of claim 6, wherein:

the first member of the set of words or phrases is associated with a second member of the set with which the first member is acoustically confusable and that, when substituted for the first member in a recognition result, changes a medical meaning of the recognition result; and

the determining whether the second recognition result includes the at least one other member associated with the first member of the set comprises determining whether the second recognition result includes the second member of the set.

8. The method of claim 1, wherein:

the results of the recognition by the ASR system comprise a plurality of results identified by the ASR system as likely to be an accurate recognition result for the speech input, and

the method further comprises selecting, from the plurality of results, the two or more results to be evaluated using the at least one criterion, the two or more results being fewer than all of the plurality of results.

9. The method of claim 1, wherein the triggering an alert concerning one of the two or more results comprises presenting a visual and/or audible message.

10. The method of claim 1, wherein the evaluating the two or more results identified by the ASR system using the at least one criterion comprises evaluating an N best list of recognition results identified by the ASR system without identifying a new order of recognition results in the N best list.

11. The method of claim 1, further comprising:

evaluating prosody information and/or information indicating the presence of one or more hesitation vocalizations, produced by the ASR system based on the speech input, to determine whether a speaker exhibited one or more signs of uncertainty when providing the speech input; and

in response to determining that the speaker exhibited one or more signs of uncertainty, triggering an alert concerning the two or more results.

12. At least one computer-readable storage medium having encoded thereon computer-executable instructions that, when executed by at least one computer, cause the at least one computer to carry out a method of processing results of a recognition by an automatic speech recognition (ASR) system on a speech input, the results comprising two or more results identified by the ASR system as likely to be an accurate recognition result for the speech input, the method comprising:

evaluating the two or more results using at least one criterion that differs from criteria used by the ASR system in determining the two or more results; and

in response to determining that the at least one criterion is met by the two or more results, triggering an alert concerning one of the two or more results.

13. The at least one computer-readable storage medium of claim 12, wherein:

the two or more results comprise a first recognition result identified by the ASR system as most likely to be a correct recognition result for the speech input, and

the evaluating the two or more results using the at least one criterion comprises determining whether the two or more results comprise an indication of a potential error in the first recognition result that may cause a meaning of the first recognition result to differ from a meaning of the speech input.

14. The at least one computer-readable storage medium of claim 13, wherein:

the two or more results comprise a first recognition result and a second recognition result;

the method further comprises semantically interpreting each of the first recognition result and the second recognition result to determine at least one first fact expressed in the first recognition result and at least one second fact expressed in the second recognition result; and

.

the determining whether the two or more results comprise an indication of a potential error that may cause the meaning of the first recognition result to differ from the meaning of the speech input comprises determining whether there is a difference between the at least one first fact and the at least one second fact.

15. A method of processing a result of a recognition by an automatic speech recognition (ASR) system on a speech input, the method comprising:

determining whether the result from the ASR system comprises a word or phrase that is unlikely to occur in a domain to which the speech input relates; and

in response to determining that the result comprises a word or phrase that is unlikely to occur in the domain to which the speech input relates, triggering an alert for the result.

16. The method of claim 15, wherein the determining whether the result from the ASR system comprises a word or phrase that is unlikely to occur in the domain comprises:

comparing at least one word or phrase in the result to a set of words and/or phrases that includes words and/or phrase that are unlikely to appear in the domain to determine whether at least one word or phrase in the result appears in the set.

17. The method of claim 15, wherein the determining whether the result comprises a word that is unlikely to occur in the domain comprises evaluating a word or phrase of the result using a language model for the domain to determine a likelihood of the word or phrase of the result occurring in the domain.

18. The method of claim 15, further comprising:

determining the domain to which the speech input relates based on information identifying the speech input.

19. The method of claim 15, wherein:

the speech input relates to a plurality of domains, and

the determining whether the result from the ASR system comprises a word or phrase that is unlikely to occur in a domain to which the speech input relates comprises;

determining a set of words and/or phrases unlikely to occur in at least one of the plurality of domains to which the speech input relates, and

determining whether the result includes a word or phrase in the set.

20. The method of claim 15, wherein the determining whether the result comprises a word or phrase that is unlikely to occur in the domain comprises determining whether the result comprises a word or phrase that is unlikely to occur in speech input relating to a medical domain.