Speech recognition error diagnosis

Patent metadata

Field	Value
Publication number	US-10019984-B2
Application number	US-201514634714-A
Country	US
Kind code	B2
Filing date	Feb 27, 2015
Priority date	Feb 27, 2015
Publication date	Jul 10, 2018
Grant date	Jul 10, 2018

Abstract

Techniques and technologies for diagnosing speech recognition errors are described. In an example implementation, a system for diagnosing speech recognition errors may include an error detection module configured to determine that a speech recognition result is least partially erroneous, and a recognition error diagnostics module. The recognition error diagnostics module may be configured to (a) perform a first error analysis of the at least partially erroneous speech recognition result to provide a first error analysis result; (b) perform a second error analysis of the at least partially erroneous speech recognition result to provide a second error analysis result; and (c) determine at least one category of recognition error associated with the at least partially erroneous speech recognition result based on a combination of the first error analysis result and the second error analysis result.

First claim

What is claimed is:

1. A system for diagnosing speech recognition errors, comprising:
at least one processing component; and
one or more media operably coupled to the at least one processing component and bearing one or more instructions that, when executed by the at least one processing component, perform operations including at least:
determine that a speech recognition result is at least partially erroneous;
perform a first error analysis of the at least partially erroneous speech recognition result to provide a first error analysis result;
perform a second error analysis of the at least partially erroneous speech recognition result to provide a second error analysis result; and
determine at least one category of recognition error associated with the at least partially erroneous speech recognition result based on a combination of the first error analysis result and the second error analysis result, including determine that the at least one category of recognition error includes at least an acoustic model error when
(a) the first error analysis result indicates that a reference language model score associated with a reference speech is higher than a recognition language model score associated with the at least partially erroneous speech recognition result; and
(b) the second error analysis result indicates that a reference acoustic model score associated with the reference speech is lower than a recognition acoustic model score associated with the at least partially erroneous speech recognition result;

determine at least one corrective action to at least partially correct at least one aspect of a speech recognition component based at least partially on the at least one category of recognition error associated with the at least partially erroneous speech recognition result; and
at least one of:
provide an indication of the at least one corrective action; or
adjust at least one aspect of the speech recognition component based on the at least one corrective action.

2. The system of claim 1 , wherein:
the first error analysis includes at least one language model scoring operation; and
the second error analysis includes at least one acoustic model scoring operation.

3. The system of claim 1 , wherein:
the first error analysis includes at least one dictionary check operation; and
the second error analysis includes at least one transcription analysis operation.

4. The system of claim 1 , wherein:
the first error analysis includes at least one emulation operation; and
the second error analysis includes at least one grammar analysis operation.

5. The system of claim 1 , wherein:
the first error analysis of the at least partially erroneous speech recognition result includes a comparison of a language model score associated with the at least partially erroneous speech recognition result with a language model score associated with a reference speech recognition result; and
the second error analysis of the at least partially erroneous speech recognition result includes a comparison of an acoustic model score associated with the at least partially erroneous speech recognition result with an acoustic model score associated with the reference speech recognition result.

6. The system of claim 1 , wherein at least one of the first error analysis or the second error analysis comprises:
one or more emulation operations that assume an ideal operation of an acoustic model to assess an actual operation of a language model.

7. The system of claim 1 , wherein the operations further comprise:
perform a third error analysis of the at least partially erroneous speech recognition result to provide a third error analysis result; and
determine at least one category of recognition error associated with the at least partially erroneous speech recognition result based on a combination of at least the first error analysis result, the second error analysis result, and the third error analysis result.

8. The system of claim 7 , wherein:
the first error analysis includes at least one language model scoring operation;
the second error analysis includes at least one acoustic model scoring operation; and
the third error analysis includes at least one of an engine setting check operation, a penalty model setting check operation, a force alignment operation, a 1:1 alignment test operation, an emulation operation, or a dictionary check operation.

9. The system of claim 1 , wherein determine at least one corrective action to at least partially correct at least one aspect of a speech recognition component based at least partially on the at least one category of recognition error associated with the at least partially erroneous speech recognition result comprises:
determine at least one corrective action to at least partially correct at least one aspect of at least one of a language model, an acoustic model, a transcription model, a pruning model, a penalty model, or a grammar of a speech recognition component based at least partially on the at least one category of recognition error associated with the at least partially erroneous speech recognition result.

10. The system of claim 1 , wherein provide an indication of the at least one corrective action comprises:
provide at least one recommended action to at least partially correct at least one aspect of at least one of a language model, an acoustic model, a transcription model, a pruning model, a penalty model, or a grammar of the speech recognition component based at least partially on the at least one category of recognition error associated with the at least partially erroneous speech recognition result.

11. The system of claim 1 , wherein adjust at least one aspect of the speech recognition component based on the at least one corrective action comprises:
adjust at least one aspect of at least one of a language model, an acoustic model, a transcription model, a pruning model, a penalty model, or a grammar of a speech recognition component based at least partially on the at least one category of recognition error associated with the at least partially erroneous speech recognition result.

12. A system for diagnosing speech recognition errors, comprising:
at least one processing component; and
one or more media operably coupled to the at least one processing component and bearing one or more instructions that, when executed by the at least one processing component, perform operations including at least:
determine that a speech recognition result is at least partially erroneous;
perform a first error analysis of the at least partially erroneous speech recognition result to provide a first error analysis result;
perform a second error analysis of the at least partially erroneous speech recognition result to provide a second error analysis result;
determine at least one category of recognition error associated with the at least partially erroneous speech recognition result based on a combination of the first error analysis result and the second error analysis result, including determine that the at least one category of recognition error includes at least an acoustic model error and a language model error when
(a) the first error analysis result indicates that a reference language model score associated with a reference speech is lower than a recognition language model score associated with the at least partially erroneous speech recognition result; and
(b) the second error analysis result indicates that a reference acoustic model score associated with the reference speech is lower than a recognition acoustic model score associated with the at least partially erroneous speech recognition result;

determine at least one

Assignees

• Microsoft Technology Licensing Llc

Inventors

• Kuo Shiun-Zu
• Reutter Thomas
• Gong Yifan
• Hanson Mark T
• Tian Ye
• Chang Shuangyu
• Hamaker Jonathan
• Miao Qi
• Tu Yuancheng

Classifications

• G10L15/01 Primary

  Assessment or evaluation of speech recognition systems · CPC title

• G10L15/183

  using context dependencies, e.g. language models · CPC title