Apparatus and method for decoding to recognize speech using a third speech recognizer based on first and second recognizer results

• US 10,115,394 B2
• Filed: 07/08/2014
• Issued: 10/30/2018
• Est. Priority Date: 07/08/2014
• Status: Active Grant

• Current Assignee: Mitsubishi Electric Corporation
• Original Assignee: Mitsubishi Electric Corporation
• Inventors: Sugitani, Naoya, Okato, Yohei, Yamazaki, Michihiro
• Primary Examiner(s): WOZNIAK, JAMES S
• Application Number:
  US15/302,576
• Publication Number:
  US 20170140752A1
• Time in Patent Office: 1,575 Days
• Field of Search: 704251, 7042701
• CPC Class Codes: G10L 15/01 Assessment or evaluation of speech recognition systems
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 15/10 using distance or distortion measures between unknown speech and reference templates
• CPC Class Codes: G10L 15/32 Multiple recognisers used in sequence or in parallel; Score combination systems therefor, e.g. voting systems

Abstract

An object is to provide a technique which can provide a highly valid recognition result while preventing unnecessary processing. A voice recognition device includes first to third voice recognition units, and a control unit. When it is decided based on recognition results obtained by the first and second voice recognition units to cause the third voice recognition unit to recognize an input voice, the control unit causes the third voice recognition unit to recognize the input voice by using a dictionary including a candidate character string obtained by at least one of the first and second voice recognition units.

Claims

1. A voice recognition apparatus which performs recognition of voice to be outputted from an output unit, the voice recognition apparatus comprising:

a processor configured to control;

first, second and third voice recognizers which each recognize an input voice and obtain a recognition result including a candidate character string corresponding to the input voice, each of said first, second and third voice recognizers include a memory that stores a dictionary; and

a controller which, when it is decided based on said recognition result obtained by each of said first and second voice recognizers to cause said third voice recognizer to recognize said input voice, causes said third voice recognizer to recognize said input voice by using the dictionary included in said third voice recognizer including said candidate character string obtained by at least one of said first and second voice recognizers, and causes said output unit to output said recognition result obtained by said third voice recognizer, wherein

the recognition results obtained by each of said first and second voice recognizers further include score values indicating accuracy of said candidate character strings, and

whether or not to cause the third voice recognizer to recognize said input voice is decided based on an index including at least one of said score values which are obtained by said first and second voice recognizers and are a maximum, a similarity indicating a degree that said candidate character strings obtained by said first and second voice recognizers match each other, and an order distance indicating a degree of difference in an order of said candidate character strings aligned in order of said score values obtained by said first and second voice recognizers.

2. The voice recognition apparatus according to claim 1, wherein,

when deciding based on said recognition result obtained by each of said first and second voice recognizers not to cause said third voice recognizer to recognize said input voice,

said controller causes said output unit to output said recognition result obtained by one of said first and second voice recognizers.

3. The voice recognition apparatus according to claim 1, wherein

said third voice recognizer

recognizes said input voice by using a dictionary unique to said third voice recognizer, together with a dictionary including said candidate character string.

4. The voice recognition apparatus according to claim 1, wherein,

in a first case where said recognition result obtained by each of said first and second voice recognizers do not completely match to each other, and said similarity is not smaller than a predetermined threshold, it is decided to cause said third voice recognizer to recognize said input voice, and, in a second case other than the first case, it is decided not to cause said third voice recognizer to recognize said input voice.

5. The voice recognition apparatus according to claim 1, wherein

said index is said similarity, and

in a first case where said recognition results obtained by said first and second voice recognizers do not completely match each other, and said similarity is not smaller than a predetermined threshold, it is decided to cause said third voice recognizer to recognize said input voice, and, in a second case other than the first case, it is decided not to cause said third voice recognizer to recognize said input voice.

6. The voice recognition apparatus according to claim 1, wherein

said index is said order distance, and

in a first case where said recognition results obtained by said first and second voice recognizers do not completely match each other and said order distance is not larger than a predetermined threshold, it is decided to cause said third voice recognizer to recognize said input voice, and, in a second case other than the first case, it is decided not to cause said third voice recognizer to recognize said input voice.

7. The voice recognition apparatus according to claim 1, wherein

said index is said score value which is maximum, and

n a first case where said recognition results obtained by said first and second voice recognizers do not completely match each other, and both of first and second score values which are maximum and are obtained by said first and second voice recognizers are smaller than predetermined first and second thresholds or are larger than said predetermined first and second thresholds, it is decided to cause said third voice recognizer to recognize said input voice, and, in a second case other than the first case, it is decided not to cause said third voice recognizer to recognize said input voice.

8. The voice recognition apparatus according to claim 1, wherein,

every time said third voice recognizer recognizes said input voice, said candidate candidate character string from each of the first and second voice recognizers for the recognition is deleted from said dictionary.

9. A voice recognition method for performing recognition of voice to be outputted from an output unit, the voice recognition method comprising,

when it is decided based on a recognition result obtained by each of first and second voice recognizers among first, second and third voice recognizers which each recognize an input voice and obtain said recognition result including a candidate character string corresponding to the input voice, to cause said third voice recognizer to recognize said input voice, causing said third voice recognizer to recognize said input voice by using a dictionary including said candidate character string obtained by at least one of said first and second voice recognizers, and

causing said output unit to output said recognition result obtained by said third voice recognizer, wherein

the recognition results obtained by each of said first and second voice recognizers include score values indicating accuracy of said candidate character strings, and

whether or not to cause the third voice recognizer to recognize said input voice is decided based on an index including at least one of said score values which are obtained by said first and second voice recognizers and are a maximum, a similarity indicating a degree that said candidate character strings obtained by said first and second voice recognizers match each other, and an order distance indicating a degree of difference in an order of said candidate character strings aligned in order of said score values obtained by said first and second voice recognizers.