Information processing apparatus, information processing method, and program

• US 8,566,094 B2
• Filed: 08/10/2011
• Issued: 10/22/2013
• Est. Priority Date: 08/26/2010
• Status: Expired due to Fees

• Current Assignee: Sony Corporation (Sony Group Corp.)
• Original Assignee: Sony Corporation (Sony Group Corp.)
• Inventors: Minamino, Katsuki, Hiroe, Atsuo, Maeda, Yoshinori, Asakawa, Satoshi
• Primary Examiner(s): Dorvil, Richemond
• Assistant Examiner(s): ESTES, ERNEST LANGSTON
• Application Number:
  US13/206,631
• Publication Number:
  US 20120053942A1
• Time in Patent Office: 804 Days
• Field of Search: None
• US Class Current: 704/251
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams
• CPC Class Codes: G10L 15/19 Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules
• CPC Class Codes: G10L 15/32 Multiple recognisers used in sequence or in parallel; Score combination systems therefor, e.g. voting systems

Abstract

An apparatus, method and program for performing a speech recognition process utilizing contextual information that comprises an estimation of the intention of an utterance of a user. The recognition process includes calculating a pre-score based on observed contextual information according intention models which correspond to a plurality of types of intention information and combining the pre-scoring results with acoustic and linguistic scores to obtain an improved recognition or comprehension of the intent of a user utterance.

Claims

1. An information processing apparatus comprising a microprocessor comprising a plurality of portions configured to be executed by the microprocessor, comprising:

a pre-score adjustment portion which calculates a pre-score based on context information obtained as observation information, for an intention model as a unit corresponding to each of a plurality of types of intention information registered in advance;

a multi-matching portion which determines the most suitable word group for an input speech based on a user'"'"'s utterance and calculates an acoustic score and a linguistic score to be given to the word group for the intention model as a unit; and

an intention determination portion which determines intention information corresponding to an intention model achieving the highest total score as an intention corresponding to the user'"'"'s utterance by comparing total scores calculated from the pre-score, the acoustic score, and the linguistic score of the intention model as a unit,

a pre-score storing portion in which a pre-score corresponding to a context with respect to each context information corresponding to a plurality of different types of observation information has been registered, wherein

the observation information includes the plurality of types of observation information,

the pre-score adjustment portion selects the pre-score corresponding to a context that has been registered in the pre-score storing portion based on the context information and calculates a pre-score for the intention model as a unit by applying the selected pre-score corresponding to a context, and

the context information as the observation information includes at least information of speech input person identification input from an image processing portion.

2. The apparatus according to claim 1,

wherein the context information as the observation information includes information of speech input microphone identification input from a microphone determination portion, and

the pre-score adjustment portion selects the pre-score corresponding to a context that has been registered in the pre-score storing portion based on the context information and calculates the pre-score for the intention model as a unit by applying the selected pre-score corresponding to the context.

3. The apparatus according to claim 2,

wherein

the pre-score adjustment portion selects the pre-score corresponding to a context which corresponds to a process category selected by the user and calculates the pre-score for the intention model as a unit by applying the selected pre-score corresponding to a context.

4. The apparatus according to claim 2,

wherein the information of speech input person identification is an identified result of a face identification process executed by the image processing portion based on an image captured by a camera of the information processing apparatus, and

the pre-score adjustment portion selects the pre-score corresponding to a context which corresponds to the speech input person identified, and calculates the pre-score for the intention model as a unit by applying the selected pre-score corresponding to a context.

5. The apparatus according to claim 2,

wherein the information of speech input microphone identification is information of a microphone determined to be a microphone from which the speech has been input, which is obtained by the microphone determination portion of the information processing apparatus, and

the pre-score adjustment portion selects the pre-score corresponding to a context which corresponds to the microphone from which the speech has been input, and calculates the pre-score for the intention model as a unit by applying the selected pre-score corresponding to a context.

6. The apparatus according to claim 1,

wherein the intention determination portion calculates a total score for the intention model as a unit by applying pre-set weight corresponding to each of the acoustic score, linguistic score, and pre-score.

7. The apparatus according to claim 1, further comprising:

a pre-score learning portion which receives the input of the determination result of the pre-score determination portion and updates the pre-score corresponding to a context that has been registered in the pre-score storing portion by a learning process performed based on input determination results.

8. An information processing method executed in an information processing apparatus comprising a microprocessor, the method comprising:

causing a pre-score adjustment portion to calculate a pre-score based on context information obtained as observation information, for an intention model as a unit corresponding to each of a plurality of types of intention information registered in advance;

causing a multi-matching portion to determine the most suitable word group for an input speech based on a user'"'"'s utterance and to calculate an acoustic score and a linguistic score to be given to the word group for the intention model as a unit;

causing an intention determination portion to determine intention information corresponding to an intention model achieving the highest total score as an intention corresponding to the user'"'"'s utterance by comparing the total scores calculated from the pre-score, acoustic score, and the linguistic score of the intention model as a unit; and

causing a pre-score storing portion to register a pre-score corresponding to a context with respect to each context information corresponding to a plurality of different types of observation information, wherein

the observation information includes the plurality of types of observation information,

causing the pre-score storing portion to register further comprises causing the pre-score storing portion to select the pre-score corresponding to a context that has been registered in the pre-score storing portion based on the context information and calculates a pre-score for the intention model as a unit by applying the selected pre-score corresponding to a context, and

the context information as the observation information includes at least information of speech input person identification input from an image processing portion.

9. A non-transitory, computer readable medium configured to cause an information processing apparatus to perform information processing comprising:

causing a pre-score adjustment portion to calculate a pre-score based on context information obtained as observation information, for an intention model as a unit corresponding to each of a plurality of types of intention information registered in advance;

causing a multi-matching portion to determine the most suitable word group for an input speech based on a user'"'"'s utterance and to calculate an acoustic score and a linguistic score to be given to the word group for the intention model as a unit;

causing an intention determination portion to determine intention information corresponding to an intention model achieving the highest total score as an intention corresponding to the user'"'"'s utterance by comparing the total scores calculated from the pre-score, the acoustic score, and the linguistic score of the intention model as a unit; and

causing a pre-score storing portion to register a pre-score corresponding to a context with respect to each context information corresponding to a plurality of different types of observation information, wherein

the observation information includes the plurality of types of observation information,

causing the pre-score storing portion to register further comprises causing the pre-score storing portion to select the pre-score corresponding to a context that has been registered in the pre-score storing portion based on the context information and calculates a pre-score for the intention model as a unit by applying the selected pre-score corresponding to a context, and

the context information as the observation information includes at least information of speech input person identification input from an image processing portion.