Language model creation device

• US 9,043,209 B2
• Filed: 09/03/2009
• Issued: 05/26/2015
• Est. Priority Date: 11/28/2008
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Yamamoto, Hitoshi, Hanazawa, Ken, Miki, Kiyokazu
• Primary Examiner(s): YEN, ERIC L
• Application Number:
  US13/129,721
• Publication Number:
  US 20110231183A1
• Time in Patent Office: 2,091 Days
• Field of Search: 704/256.2, 704/257
• US Class Current: 704/256.2
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/19 Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams

Abstract

This device 301 stores a first content-specific language model representing a probability that a specific word appears in a word sequence representing a first content, and a second content-specific language model representing a probability that the specific word appears in a word sequence representing a second content. Based on a first probability parameter representing a probability that a content represented by a target word sequence included in a speech recognition hypothesis generated by a speech recognition process of recognizing a word sequence corresponding to a speech, a second probability parameter representing a probability that the content represented by the target word sequence is a second content, the first content-specific language model and the second content-specific language model, the device creates a language model representing a probability that the specific word appears in a word sequence corresponding to a part corresponding to the target word sequence of the speech.

Claims

1. A language model creation device comprising:

a language model creating unit configured to execute a language model creation process of;

acquiring a first content-specific language model which represents an appearance probability that a specific word appears in a first content, the first content comprising a first word sequence, a second content-specific language model which represents an appearance probability that the specific word appears in a second content, the second content comprising a second word sequence, a first probability parameter representing a probability that a content represented by a target word sequence is the first content, and a second probability parameter representing a probability that the content represented by the target word sequence is the second content, the target word sequence being at least a part of a speech recognition hypothesis generated in a speech recognition process; and

creating a language model based on the first probability parameter, the second probability parameter, the first content-specific language model and the second content-specific language model, the created language model representing a combined appearance probability which is a probability that the specific word appears within at least a portion of the target word sequence.

2. The language model creation device according to claim 1, wherein:

the language model creating unit is configured to create the language model such that the combined appearance probability increases as a sum of (i) a product of a first coefficient and the probability represented by the first content-specific language model, and (ii) a product of a second coefficient and the probability represented by the second content-specific language model, becomes larger,

wherein the first coefficient increases in value as the first probability parameter becomes larger, and the second coefficient increases in value as the second probability parameter becomes larger.

3. The language model creation device according to claim 1, wherein:

the language model creating unit is configured to execute the language model creation process on each of a plurality of target word sequences obtained by dividing the speech recognition hypothesis on a border of words.

4. The language model creation device according to claim 1, wherein:

the language model creating unit is configured to execute the language model creation process on each of a plurality of target word sequences obtained by dividing the speech recognition hypothesis at a position different from a border of words.

5. The language model creation device according to claim 3, wherein:

the language model creating unit is configured to, in a case that both the first probability parameter and the second probability parameter are smaller than a preset lower limit threshold, set a value of the first probability parameter and a value of the second probability parameter equal to each other.

6. The language model creation device according to claim 3, wherein:

the language model creating unit is configured to, in a case that the first probability parameter, acquired with respect to a certain target word sequence of the plurality of target word sequences, is larger than a preset upper limit threshold, increase the first probability parameter, acquired with respect to an adjacent target word sequence which is adjacent to the certain target word sequence.

7. The language model creation device according to claim 1, wherein:

the language model creating unit is configured to acquire the first probability parameter and the second probability parameter, based on a theory of conditional random fields.

8. The language model creation device according to claim 1, wherein:

the language model creating unit is configured to acquire the first probability parameter and the second probability parameter, based on at least one of word attribute information and speech recognition process information, the word attribute information representing an attribute of a word of the target word sequence, the speech recognition process information being acquired when the speech recognition process is executed.

9. The language model creation device according to claim 8, wherein:

the word attribute information includes at least one of information representing a word surface, information representing pronunciation, and information representing a part of speech.

10. The language model creation device according to claim 8, wherein:

the speech recognition process information includes at least one of information representing recognition reliability, information representing a length of duration, and information representing presence/absence of a preceding silence, the recognition reliability being reliability of a recognition result of the speech recognition process, the length of duration being a time period during which one sound continues.

11. The language model creation device according to claim 1, wherein:

the language model creating unit is configured to acquire, based on at least one of information representing a position where the target word sequence is in the speech recognition hypothesis, the first probability parameter and the second probability parameter, information representing that the speech recognition hypothesis includes a plurality of same words, information representing a contents'"'"' connective state in the speech recognition hypothesis, and information representing, in a case that there are a plurality of speech recognition hypotheses, a relation between the speech recognition hypotheses.

12. A speech recognition device, comprising:

a language model creating unit configured to execute a language model creation process of;

acquiring a first content-specific language model which represents an appearance probability that a specific word appears in a first content, the first content comprising a first word sequence, a second content-specific language model which represents an appearance probability that the specific word appears in a second content, the second content comprising a second word sequence, a first probability parameter representing a probability that a content represented by a target word sequence is the first content, and a second probability parameter representing a probability that the content represented by the target word sequence is the second content, the target word sequence being at least a part of a speech recognition hypothesis generated in a speech recognition process; and

creating a language model based on the first probability parameter, the second probability parameter, the first content-specific language model and the second content-specific language model, the created language model representing a combined appearance probability which is a probability that the specific word appears within at least a portion of the target word sequence; and

a speech recognizing unit configured to execute a speech recognition process of recognizing a word sequence corresponding to an inputted speech, based on the language model created by the language model creating unit.

13. The speech recognition device according to claim 12, wherein:

the language model creating unit is configured to create the language model such that the combined appearance probability increases as a sum of (i) a product of a first coefficient and the probability represented by the first content-specific language model, and (ii) a product of a second coefficient and the probability represented by the second content-specific language model, becomes larger,

wherein the first coefficient increases in value as the first probability parameter becomes larger, and the second coefficient increases in value as the second probability parameter becomes larger.

14. The speech recognition device according to claim 12, wherein:

the speech recognizing unit is configured to execute the speech recognition process and thereby generate the speech recognition hypothesis;

the language model creating unit is configured to create the language model, based on the speech recognition hypothesis generated by the speech recognizing unit; and

the speech recognizing unit is configured to again execute the speech recognition process, based on the language model created by the speech model creating unit.

15. The speech recognition device according to claim 12, configured to execute an iterative process of alternately and repeatedly executing the speech recognition process and the language model creation process, the speech recognition process being executed, based on each language model created by the language model creating unit, by the speech recognizing unit, the language model creation process being executed, based on each word sequence recognized by the speech recognizing unit, by the language model creating unit.

16. The speech recognition device according to claim 15, configured to finish the iterative process in a case that a predetermined termination condition is satisfied.

17. The speech recognition device according to claim 16, wherein:

the termination condition is a condition that a word sequence recognized in a previous recognition process coincides with a word sequence recognized in a most recent speech recognition process.

18. The speech recognition device according to claim 16, wherein:

the termination condition is a condition that a number of times of execution of the speech recognition process is larger than a preset threshold number of times.

19. A language model creation method executed by at least one processor, comprising:

acquiring a first content-specific language model which represents an appearance probability that a specific word appears in a first content, the first content comprising a first word sequence, a second content-specific language model which represents an appearance probability that the specific word appears in a second content, the second content comprising a second word sequence, a first probability parameter representing a probability that a content represented by a target word sequence is the first content, and a second probability parameter representing a probability that the content represented by the target word sequence is the second content, the target word sequence being at least a part of a speech recognition hypothesis generated in a speech recognition process; and

creating, by the at least one processor, a language model based on the first probability parameter, the second probability parameter, the first content-specific language model and the second content-specific language model, the created language model representing a combined appearance probability which is a probability that the specific word appears within at least a portion of the target word sequence.

20. The language model creation method according to claim 19, comprising:

creating the language model such that the combined appearance probability increases as a sum of (i) a product of a first coefficient and the probability represented by the first content-specific language model, and (ii) a product of a second coefficient and the probability represented by the second content-specific language model, becomes larger,

wherein the first coefficient increases in value as the first probability parameter becomes larger, the second coefficient increases in value as the second probability parameter becomes larger.

21. A non-transitory computer-readable medium storing a language model creation program comprising instructions for causing an information processing device to realize:

a language model creating unit configured to execute a language model creation process of;

acquiring a first content-specific language model which represents an appearance probability that a specific word appears in a first content, the first content comprising a first word sequence, a second content-specific language model which represents an appearance probability that the specific word appears in a second content, the second content comprising a second word sequence, a first probability parameter representing a probability that a content represented by a target word sequence is the first content, and a second probability parameter representing a probability that the content represented by the target word sequence is the second content, the target word sequence being at least a part of a speech recognition hypothesis generated in a speech recognition process; and

creating a language model based on the first probability parameter, the second probability parameter, the first content-specific language model and the second content-specific language model, the created language model representing a combined appearance probability which is a probability that the specific word appears within at least a portion of the target word sequence.

22. The non-transitory computer-readable medium storing the language model creation program according to claim 21, wherein:

the language model creating unit is configured to create the language model such that the combined appearance probability increases as a sum of (i) a product of a first coefficient and the probability represented by the first content-specific language model, and (ii) a product of a second coefficient and the probability represented by the second content-specific language model, becomes larger,

wherein the first coefficient increases in value as the first probability parameter becomes larger, the second coefficient increases in value as the second probability parameter becomes larger.