JP2013137460A - Speech recognition device, method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a speech recognition device, method and program having high speech recognition performance.SOLUTION: A speech recognition device comprises: a speech recognition section 10 for recognizing user's speech using a language model; an independent word extraction section 21 for extracting an independent word included in the user's speech; an independent word relevance DB 31 for storing an independent word relevant to each other in association with relevance; independent word characteristic DB 32 for storing characteristic information of an independent word included in a text used by a user; a priority calculation section 22 for calculating priority of a relevant independent word relevant to the independent word extracted by the independent word extraction section 21; and a language model updating processing section 24 for adjusting weight of the language model in accordance with the priority.

Description

  The present invention relates to a speech recognition apparatus, method, and program, and more particularly to a speech recognition apparatus, method, and program for performing speech recognition using a language model.

  In recent years, a speech recognition device that recognizes speech spoken by a speaker has been used (Patent Document 1). In Patent Document 1, a keyword is stored, and the language model score is adjusted to increase the keyword recognition rate. However, in Patent Document 1, there is a problem in the ease of recognizing a keyword immediately after word registration.

JP 2010-197411 A

  In such speech recognition, an important language model for speech recognition of free sentences is created using a text corpus that is available in advance. Also, the topic and the corresponding person are easily changed according to the TPO. For example, while talking about the topic of Okinawa travel, “I ate a lot of soki soba” became “I ate a lot of soba”. The model of “early” + “soba” is more likely to appear than the word “soki soba”. It is difficult to maintain sufficient speech recognition performance with only a general language model.

For example, the N-gram language model is a model that is calculated based on conditional probabilities as shown in the following expression, where P (w) is the appearance probability of an input word.
P (w) = P (w _i | w _i−N−1 ... W _i−1 )

In the N-gram language model, the generation probability of the i-th word w _i depends on (N−1) word strings w _i−N−1 ... W _i−2 w _i−1 . For example, taking 3-gram (trigram) as an example, the probability that the word w ₃ appears following the word string w ₁ w ₂ is P (w ₃ | w ₁ w ₂ ).

  The learning data for obtaining the conditional probability is performed using a corpus such as a newspaper or web information. If you select solid Japanese-like information, there are a lot of hard tone and news, and it is difficult to maintain sufficient speech recognition performance with a general language model alone in a wide range of tasks such as chatting. In Patent Document 1, it is necessary to input a keyword. In addition, if the topic changes greatly, the input keyword is not related to the topic that is actually spoken. Therefore, an appropriate language model cannot be updated and sufficient speech recognition performance may not be obtained.

  The present invention has been made to solve such a problem, and an object thereof is to provide a speech recognition apparatus, method, and program having high speech recognition performance.

  A speech recognition apparatus according to an aspect of the present invention includes a speech recognition unit that recognizes a user's speech using a language model, an extraction unit that extracts an independent word included in the user's speech, and an independent word associated with each other. Using the degree-of-association storage means for storing the degree-of-association, the independent word characteristic storage means for storing the characteristic information of the independent words included in the sentence used by the user, the characteristic information, and the degree of association, Priority calculating means for calculating priorities of related independent words related to the independent words extracted by the extracting means, and adjusting means for adjusting the weight of the language model according to the priorities. Is. According to this configuration, the relevance storage means is referred to, and the priority is calculated using the independent words extracted from the speech. Furthermore, the priority is calculated using the correction information included in the recognition result. For this reason, the weight of a language model can be adjusted appropriately and speech recognition performance can be improved.

  In the above speech recognition apparatus, the independent word characteristic storage means stores at least one of time information of the independent word, position information of the independent word, and user information of a user who uses the independent word as the characteristic information. You may do it. By doing so, the priority can be calculated appropriately.

  The speech recognition apparatus further includes a determination unit that determines whether or not the related independent word is a priority word according to the priority of the related independent word, and the adjustment unit includes a score related to the priority word. The language model may be updated so as to be relatively large. In this configuration, since it is determined whether or not it is a priority word according to the priority and the score of the priority word is relatively increased, the speech recognition performance can be improved.

  The speech recognition apparatus calculates a correction coefficient based on a comparison result between the characteristic information and the current information, and a plurality of independent words stored in the association degree storage unit are extracted by the extraction unit. The priority is calculated by multiplying the sum of the degrees of association with the independent word by the correction coefficient, and the determining means determines that the related independent word is the one according to the comparison result between the priority and the threshold. It may be determined whether or not it is a priority word. Thereby, since an appropriate priority word can be extracted, speech recognition performance can be improved.

  The speech recognition apparatus calculates a correction coefficient based on a comparison result between the characteristic information and the current information of the independent words extracted by the extraction unit, and the independent words stored in the association degree storage unit. The priority is calculated by multiplying the sum of the degrees of association with a plurality of independent words extracted by the extraction means by the correction coefficient, and the determination means includes the priority among the related independent words. The upper N (N is a natural number) independent words may be determined as the priority words. Thereby, since an appropriate number of priority words can be extracted, speech recognition performance can be improved.

  In the above speech recognition apparatus, the independent words may be limited to nouns, adjectives, and verbs. Thereby, an appropriate self-supporting word can be extracted.

  The speech recognition apparatus according to any one of claims 1 to 5, wherein the relevance is set according to a co-occurrence frequency of two independent words in the sentence. . Thereby, the degree of association can be set appropriately.

  The speech recognition method according to an aspect of the present invention includes a step of recognizing a user's speech using a language model, a step of extracting an independent word included in the user's speech, and a degree of association storage unit. The related independent words related to the extracted independent words using the degree of association between the related independent words and the characteristic information of the independent words included in the sentence used by the user stored in the independent word characteristic storage means And calculating the priority of the language model according to the priority. In this method, since the priority is calculated using the independent words extracted from the speech while referring to the relevance storage means, the weight of the language model can be adjusted appropriately. Therefore, voice recognition performance can be improved.

  A speech recognition program according to an aspect of the present invention includes a step of recognizing a user's speech using a language model for a computer, a step of extracting an independent word included in the user's speech, and a relevance storage The degree of association between the related independent words stored in the means, and the characteristic information of the independent words included in the sentences used by the user stored in the independent word characteristic storage means, and the extracted independent words A step of calculating a priority of related related independent words, and a step of adjusting a weight of the language model according to the priority. According to this program, since the priority is calculated using the independent words extracted from the speech while referring to the relevance storage means, the weight of the language model can be adjusted appropriately. Therefore, voice recognition performance can be improved.

  According to the present invention, it is possible to provide a speech recognition apparatus, method, and program having high speech recognition performance.

1 is a block diagram showing a configuration of a speech recognition apparatus according to a first exemplary embodiment. It is a figure for demonstrating the independent word extraction process of a speech recognition apparatus. It is a figure for demonstrating the priority calculation process of a speech recognition apparatus. It is a table | surface which shows an example of the independent word relevance DB used for the priority calculation process of a speech recognition apparatus. It is a figure for demonstrating the priority discrimination | determination process of a speech recognition apparatus. It is a figure for demonstrating the update process of the language model of a speech recognition apparatus. FIG. 10 is a diagram for explaining speech recognition processing according to the second exemplary embodiment; FIG. 10 is a diagram for explaining priority calculation processing according to the second embodiment;

Embodiment 1 FIG.
Hereinafter, embodiments of the present invention will be described with reference to FIGS. 1 to 6. FIG. 1 is a block diagram showing a configuration of a speech recognition apparatus according to the present embodiment and a processing flow thereof. 2-6 is a figure for demonstrating the process of a speech recognition apparatus. The speech recognition unit 10, the acoustic model 13, the Ngram correction model 14, the recognition result history 15, the Ngram language model 16, the priority word estimation unit 20, the ontology 30, and the knowledge learning unit 40 are provided. The voice recognition unit 10 includes a feature amount extraction unit 11 and a similarity calculation unit 12. The priority word estimation unit 20 includes an independent word extraction unit 21, a priority calculation unit 22, a priority word determination unit 23, a language model update processing unit 24, and a priority correction unit 25. The ontology 30 includes an independent word relevance DB (database) 31 and an independent word characteristic DB 32. The knowledge learning unit 40 includes a user request 41, an independent word extraction unit 42, and a history management unit 43.

  A voice signal from a microphone is input to the voice recognition unit 10. The voice recognition unit 10 recognizes the input voice and outputs a voice recognition result (for example, text data). Specifically, the feature quantity extraction unit 11 performs Fourier transform on the audio data to extract the feature quantity. Then, the similarity calculation unit 12 performs similarity calculation using the acoustic model 13 and the Ngram correction model 14. For example, similarity calculation by pattern matching is performed on the feature amount pattern extracted by the feature amount extraction unit 11. By doing so, text data which is a voice recognition result is generated.

  The acoustic model 13 is a model representing what kind of sound a certain word corresponds to in order to obtain what kind of feature amount pattern (feature vector) is output with what probability. The Ngram modification model 14 is a language model obtained by updating an Ngram language model 16 described later. For example, the Ngram language model 16 is a model in which, for example, the connection between words (morphemes) is statistically calculated from a large number of sentences. When the previous word string is known, it is predicted which word will appear with a certain probability based on the language model. Based on the Ngram modification model 14, a score is given for the appearance probability that words appear connected to each other, and speech recognition is performed based on the score. As the Ngram language model 16, for example, a 3-gram (trigram) language model can be used. In addition, about the process in the speech recognition part 10, since a well-known method can be used, detailed description is abbreviate | omitted.

  Here, we explain an example where a speaker is talking about traveling in Okinawa. As shown in FIG. 2, the speech recognition unit 10 “goes to Okinawa”, “goes on a trip with family”, “swim in the sea”, “is beautiful”, “eats a lot of soki soba. Is output as a voice recognition result.

  The recognition result history 15 stores a history of speech recognition results in the speech recognition unit 10 as a database. Therefore, the above sentence is stored in the recognition result history 15. The recognition result history 15 stores text data that is a recognition result according to a time series together with the acquisition time.

Next, the priority word estimation unit 20 estimates the priority word based on the recognition result history stored in the recognition result history 15. First, the independent word extraction unit 21 extracts an independent word from the speech recognition result stored in the recognition result history 15. Here, the definition of independent words is limited to nouns, verbs, and adjectives. That is, the independent word extraction unit 21 may not extract auxiliary verbs, particles, adjective verbs, adverbs, conjunctions, conjunctions, impression verbs, and the like. In the above example, “Okinawa”, “Go”, “Family”, “Travel”, “Go”, “Sea”, “Swim”, “Pretty”, “Soki Soba”, “Eat” are extracted as independent words. The Here, a group of independent words taken out of n in the order of the independent words history from the recent ones and V _n. When referring to the history of the last 10 independent words, n = 10 and V ₁₀ = [eat, soki soba, beautiful,..., Okinawa]. Hereinafter, an example of extracting a group of 10 independent words will be described. Of course, the number of free words included in V _n may be one or more, and may overlap.

  Based on the independent words extracted by the independent word extraction unit 21, the priority calculation unit 22 calculates the priority. The priority calculation unit 22 refers to the independent word association degree DB 31 and the independent word characteristic DB 32 as the ontology 30 and calculates the priority. In the independent word relevance DB 31, two independent words (related word pairs) that are related to each other are stored in association with the relevance. That is, in the independent word relevance DB 31, independent words are converted into an ontology.

The number n of independent words extracted by the independent word extraction unit 21 is 10. In the case of a group V ₁₀ of the most recent content words are extracted, taking priority noun "Shiisa" appears in the speech as an example. The degree of association between two independent words is defined in advance in the independent word association degree DB 31. The degree of association is set as a value between 0 and 1. The degree of association between the noun “shisa” stored in the independent word association degree DB 31 and “eat” included in V ₁₀ is extracted. Similarly, other independent words included in the V _10, for example, "Soki soba", the "beautiful" and the like also extracts the relation level of a pair of the "scissor". In the independent word relevance DB 31, as shown in FIG. 3, the relevance level of the related word pair “Shisa” and “Eat” is 0.0001, and “Shisa” and “Soki Soba” The relevance level of the related word pair (Shisar | Soki soba) is 0.21, and the relevance level of the related word pair of “Shisar” and “Pretty” (Shisa | Premium) is 0.011. The degree of association is a value indicating the degree of association between the independent word extracted from the recognition result history 15 (extracted independent word) and the independent word related to the extracted independent word (related independent word), and the relationship between the two independent words The higher the is, the greater the degree of relevance. As described above, the degree of association is set according to the degree of association between two independent words.

  Here, FIG. 4 shows an example of data stored in the independent word association degree DB 31. FIG. 4 is a table showing an example of data stored in the independent word relevance DB 31. First, a self-supporting word as a key and a self-supporting word paired with the self-supporting word are stored in association with the degree of association. That is, two independent words (related word pairs) related to each other and the degree of association corresponding to the related word pair are arranged in a horizontal row. A large number of association degree pairs are registered in the independent word association degree DB 31. For example, the key independent word “Okinawa” is paired with “Shisar”, “Dugong”, “Ishigakijima” ... “Taco Rice”, and the degree of association is set for each. Yes. Similarly, independent word pairs and their degrees of association are set for independent word keys “Ishigakijima” and “Shisar”.

  Here, the degree of association is set according to the co-occurrence frequency of independent words in a large number of sentences. For example, a plurality of sentences are prepared, the number of times that two independent words are included in one sentence is counted, and the number of times is set as a co-occurrence frequency. For the pair with high co-occurrence frequency (N times or more), for example, the related word pair of “Okinawa” and “Shisar”, the related word pair of “Okinawa” and “Dugong”, etc. Shisa) and (Okinawa | Dugong) are 0.9. For a pair with a low co-occurrence frequency (less than M times), for example, a related word pair of “Shisar” and “Sango”, the relevance (Shisa | Sango) is set to 0.1. And about pairs with medium frequency of co-occurrence (more than M times and less than N times), for example, related word pairs of “Okinawa” and “US military base”, related words pair of “Okinawa” and “Taco rice” Is 0.5 (Okinawa | US Army Base) and (Okinawa | Taco Rice).

  Further, for a secondary related word of an independent word having a high degree of association (co-occurrence frequency), the degree of association is set to 0.1. For example, the related word pair of “Okinawa” and “Shisar” has a relevance level (Okinawa | Shisar) of 0.9, and the related word pair of “Okinawa” and “Dugong” has a relevance level (Okinawa | Dugon ) Is 0.9. For this reason, “Shisar” and “Dugong” are related via “Okinawa”. Therefore, for the related word pair of “Dugong” of “Shisar”, the relevance (Shisa | Dugong) is set to 0.1. For combinations that are not in the table (for example, related word pairs that have a co-occurrence frequency of 0 and are not secondary related words), the degree of association is set to 0. Of course, the setting of the independent word relevance DB 31 is not particularly limited. For example, in the above example, the degree of relevance has four levels of 0, 0.1, 0.5, and 0.9.

And the priority calculation part 22 calculates a priority based on the extracted relevance degree, and the priority correction | amendment part 25 correct | amends a priority based on the independent word relevance DB31. Specifically, the priority correction unit 25 calculates correction coefficients j _{1 to} j _l for correcting the priority based on the characteristic information included in the independent word characteristic DB 32. When the priority of the independent word w stored in the independent word relevance DB 31 is Priority (w), the sum of the relevance is obtained as shown in the following formula (1), and the correction coefficient j _{1 is} added to the sum of the relevance. The priority Priority (w) is calculated by multiplying .about.j _l .

Incidentally, w _i is the content words are independent word extraction unit 21 extracts, here is extracted ten independent words. The priority calculation unit 22 extracts the degrees of association between “Shisar” and ₁₀ independent words included in V10. The priority calculation unit 22 obtains the sum of the relevance levels for each independent word and calculates the product of the sum of the 10 relevance levels and the correction coefficient as the priority. The priority is corrected based on the characteristic information included in the independent word characteristic DB 32. Therefore, as shown below, the priority Priority of the independent word “Shisar” can be calculated.

Priority (Shisar | Eat) + (Shisar | Soki Soba) + (Shisar | Beautiful) + (Shisar | Swim) + (Shisar | Sea) + (Shisar | Go) + (Shisa | Travel) + ( Shisa | Family) + (Shisa | Go) + (Shisa | Okinawa)} × j ₁ × j ₂ ... × j _l

Here, for setting the correction coefficient _j 1 to j _l, the content words characteristics DB32 be described. The independent word characteristic DB 32 is generated by knowledge learning in the knowledge learning unit 40. For example, in the knowledge learning unit 40, the independent word extraction unit 42 extracts an independent word from the sentence used by the user request 41. And the knowledge learning part 40 produces | generates the independent word characteristic DB32 by accumulating the extracted independent word. Specifically, the independent word extraction unit 42 extracts the independent words that have appeared in the terminal device that is a voice recognition device. For example, the independent word extraction unit 42 extracts an independent word in a sentence input by a user using a terminal device such as a smartphone or a personal computer, or an independent word displayed or stored according to an input result. That is, the independent word extraction unit 42 extracts independent words included in the text used by the user.

  It is assumed that a user registers a schedule in a calendar, inputs to a memo pad, transmits an email, searches or browses the web, using a keyboard or touch panel of a terminal device. At that time, the sentence input by the user request 41 is set as a target sentence for knowledge learning. Furthermore, the sentences included in the browsed web page and the received mail are set as the target sentences for knowledge learning. Schedules registered in the calendar, input text to the memo pad, input text of the outgoing mail, text included in the received mail, web search keywords, text in the browsed web page are the target text. And the independent word extraction part 42 extracts an independent word from the target sentence by a user request. Note that the target sentence here is not limited to a complete sentence with a subject and a predicate, but includes words such as keywords.

  The independent word extraction unit 42 extracts independent words for a long time. The independent words extracted by the independent word extraction unit 42 are limited to nouns, verbs, and adjectives as described above. The history management unit 43 accumulates the extracted independent words, the number of extractions, and the like, and manages the history of the independent words. For example, the occurrence time of the extracted independent words, the number of extractions, etc. are managed in association with each other. The generation time can be a time when a user inputs a sentence or a time when a sentence data is received. Further, when the input terminal device has a position specifying function such as a GPS (Global Positioning system) function, the history management unit 43 manages the position coordinates of the place where the independent word is generated. The occurrence location can be the terminal position at the occurrence time. Further, when user information such as a user ID (user name) is registered in advance in the terminal device, the history management unit 43 manages the user information. And the independent word characteristic DB32 memorize | stores the information of the independent word characteristic managed by the log | history management part 43 as a database. That is, the independent word characteristic DB 32 stores the management result of the independent word history by the history management unit 43. Therefore, the independent word characteristic DB 32 stores the independent word characteristic information included in the sentence used by the user on the screen of the terminal device. Of course, an accumulation period for accumulating independent words may be set in advance. Of course, an accumulation period for accumulating independent words may be set in advance.

  For example, as shown in FIG. 3 and FIG. 5, an appearance in which a user having a user ID of “aaa” caused an independent word “Shisa” to appear at 12:00 in position coordinates (1234, 5678) The independent word characteristic DB 32 stores the independent word characteristic such that the number of times is 25. Furthermore, in the example of FIG. 3, the number of times that the user having the user ID “aaa” caused the independent word “buy” to appear at 12:12 in the position coordinates (3456, 7890) is 11 times. The number of appearances of a user having a user ID of “bbbb” with the independent word “shisa” appearing at 12:30 in position coordinates (1245, 5688) is three. The number of appearances is the number of appearances during the independent word accumulation period, and indicates the appearance frequency in a predetermined time. Note that the independent word characteristic DB 32 stores, for example, a history management result of independent word extraction over a long period of one year. Such independent word characteristic information may be shared on a network such as the Internet. That is, the independent word extraction unit 42 extracts independent words from sentences transmitted on the network and sentences received from the network. In this way, the knowledge learning unit 40 performs knowledge learning, and creates and updates the independent word characteristic DB 32.

Further, as shown in FIG. 5, the terminal device uses a terminal device in which the current time is 12:03, the current position is (1500, 5600), and the user whose user ID is “aaa” performs voice recognition. Suppose you are. That is, it has current information such as current time, current position, and current user ID. The current time is input to the priority correction unit 25. The priority correction unit 25 compares the independent word characteristic information stored in the independent word characteristic DB 32 with the current information, and corrects the correction coefficients j _{1 to} j _l based on the comparison result. Is calculated. Here, a specific example for obtaining the correction coefficients j _{1 to} j _l will be described.

The priority correction unit 25 calculates correction coefficients j _{1 to} j _l based on the independent word characteristic DB 32. The priority correction unit 25 determines the correction coefficient j _l based on the time information. For example, the priority correction unit 25 extracts the occurrence time and the number of appearances from the independent word characteristic DB 32 for the independent word w for which the priority is obtained. For an independent word w having an absolute value of the difference between the current time and the occurrence time within 30 minutes (within 30 before and after) and the number of appearances of N or more, j ₁ = 2.0, otherwise Let j ₁ = 1.0. That is, for a self-supporting word whose number of appearances at a certain time within a certain time before and after the current time is equal to or greater than a certain value, the priority correction unit 25 corrects the priority to be relatively large. Priority correction unit 25, the correction coefficient j ₁ when the appearance number is equal to or greater than a predetermined, may be greater than the correction coefficient j ₁ when the predetermined value or less.

Priority correction unit 25 determines the correction coefficient j ₂ on the basis of the position information. For example, the priority correction unit 25 extracts the occurrence position and the number of occurrences from the independent word characteristic DB 32 for the independent word w for which the priority is obtained. Then, j ₂ = 3.0 is set for an independent word w having a distance from the current position to the occurrence position within 100 m and M times or more of appearances, and j ₂ = 1.0 in other cases. That is, for a self-supporting word whose number of appearances at a certain position within a certain distance from the current position is equal to or greater than a certain value, the priority correction unit 25 corrects the priority to be relatively large.

Priority correction unit 25 determines the correction coefficient j ₃ based on the user information. For example, the priority correction unit 25 extracts the user ID and the number of occurrences from the independent word characteristic DB 32 for the independent word w for which the priority is obtained. Then, j ₃ = 5.0 is set for independent words whose user IDs match and the number of appearances is L or more, and j ₃ = 1.0 otherwise. That is, for the same user ID, for independent words whose appearance count is a certain value or more, the priority correction unit 25 corrects the priority to be relatively large. If user information other than the user ID is registered in the terminal device in advance, the correction coefficient may be determined using the user information. For example, the user's sex, age, address, birthplace, hobby, asset, educational background, income, etc. may be set as user information. That is, user attributes may be used as user information. Then, the priority correction unit 25 performs correction so that the priority is relatively increased for independent words whose user information matches and whose appearance count is a certain value or more.

For example, in the example using the correction coefficients j _{1 to} j ₃ , assuming that j ₁ = 1.0, j ₁ = 1.0, and j ₁ = 5.0, Priority (Shisar) = {(Shisar | Eat) + (Shisar | Soki Soba) + (Shisar | Beautiful) + (Shisar | Swim) + (Shisar | Sea) + (Shisar | Go) + (Shisar | Travel) + (Shisar | Family) + (Shisa | Go) ) + (Seaser | Okinawa)} × 1.0 × 1.0 × 5.0 = 2.1.

In the above example, the example using the correction coefficients j _{1 to} j ₃ has been described, but the number of correction coefficients is not limited to three. For example, at least one of the above correction coefficients j _{1 to} j ₃ may be used, or two or more may be used in combination. Furthermore, other characteristic information may be extracted from the independent word characteristic DB 32 and the correction coefficient may be used.

  The priority word determination unit 23 determines whether or not the related independent word is a priority word based on the priority corrected by the priority correction unit 25. The priority word determination unit 23 compares the priority with the threshold value Th, and determines whether or not the word is a priority word according to the comparison result. For example, it is assumed that the threshold value Th is set to 1.0 in the priority word determination unit 23 in advance. In the above example, Priority (Shisar) = 2.1, which is equal to or greater than the threshold Th, it is determined that “Shisar” is a priority word. Of course, the priority calculation unit 22 calculates priorities for independent words other than “Shisar” stored in the independent word relevance DB 31. Then, the priority word determination unit 23 determines whether or not the independent words other than “Shisar” are priority words according to the comparison result between the priority and the threshold value Th. It should be noted that another method may be used to determine whether or not it is a priority word. For example, the top N (N is a natural number) words with high relevance may be determined as priority words. Furthermore, the priority word may be selected by combining the determination based on the threshold Th and the determination based on the upper N words.

  Next, the language model update processing unit 24 updates the Ngram language model 16. In the Ngram language model 16, the connection between words is weighted based on many sentence examples. The language model update processing unit 24 adjusts the weight of the Ngram language model 16 related to the priority word. Thereby, priority words are recognized preferentially in the speech recognition process. Specifically, when there is an Ngram element containing a priority word, the score in the language model is updated according to a certain conversion formula. For example, when “Shisar” is determined as a priority word as described above, the score of the element including “Shisar” is increased. Based on this score, a word having the maximum appearance probability (conditional probability) is obtained. Thereby, in a language model, a priority word is weighted and the sentence containing a priority word becomes easy to be recognized. Note that the sum of the appearance probabilities may exceed 1 by increasing the above score. That is, the sum of the appearance probabilities of all words may exceed 1.

  As shown in FIG. 6, the score including “Shisar”, which is a priority word such as Ngram language model “I-ha-shisa” or “see shisa-shi”, is multiplied by ten. The other elements, that is, the elements that do not include the preferred word (here, “I-I-Lion”, “Lion-Look”, etc.) are used as they are. If the speaker is talking about Okinawa travel, independent words that are highly relevant to Okinawa are determined as preferred words. For this reason, it can be recognized as “soki soba” instead of “early soba”, and speech recognition performance can be improved. Furthermore, the priority is corrected using correction information extracted from the recognition result history 15. Thereby, a priority can be calculated appropriately and voice recognition performance can be improved.

  As described above, a conversion formula that uses score (L) = score (L) × 10 is used when a priority word is applicable. As the score conversion formula, a conversion formula score (L) = score (L) × m for multiplying m (m is a positive number) can be used. Further, as a conversion formula, a conversion formula score (L) = score (L) + a that adds a constant a (a is a positive number) may be used. Of course, the conversion equation core (L) = score (L) × m + a combining the multiple m and the constant a may be used.

  Further, the multiple m of the priority word and the constant a may be changed according to the priority value. Normalization may be performed according to the number of independent words determined as priority words. For example, when the number of independent words identified as priority words is large, the value of multiple m or constant a is reduced, and when the number of independent words identified as priority words is small, the value of multiple m or constant a The value of may be increased.

  The language model updated by the language model update processing unit 24 is referred to as an Ngram correction model 14. In this way, by updating the Ngram language model 16 as needed, it is possible to realize a speech recognition device that can be easily recognized according to the current task. Therefore, a voice recognition device with high voice recognition performance can be realized. Further, since the independent words that are highly related to the history of the speech recognition results are weighted, the speech recognition performance can be further improved. Even if the topic changes, it can respond appropriately. Furthermore, since the priority correction coefficient is calculated using the characteristic information of the independent word characteristic DB 32, the sound recognition performance can be improved. The language model may be updated for each utterance, for example.

  By calculating the correction coefficient using the time information included in the characteristic information, it is possible to change a self-recognized word that is easily recognized according to the time. For example, an independent word having a high appearance frequency in a specific time zone has a high priority in the vicinity of the time zone, and is easily discriminated as a priority word. Therefore, it becomes possible to change independent words that are easily recognized between day and night. By using the position information included in the characteristic information, it is possible to change an independent word that is easily recognized according to the utterance location. For example, an independent word having a high frequency of appearance in a specific amusement park or park is likely to be a priority word for utterances in the vicinity of the place. Furthermore, by calculating the correction coefficient using the user information included in the characteristic information, the recognized independent word can be changed according to the user ID (user name) and the user attribute. For example, when the user information matches, it is easy to distinguish the priority word. As described above, by using the characteristic information, the speech recognition performance can be further improved. Such independent word characteristic information may be shared on a network such as the Internet. For example, the server may collect information on independent word characteristics and create the independent word characteristic DB 32. Note that a terminal device (information processing device) for generating the independent word characteristic DB 32 and a terminal device that performs voice recognition may be different from each other. Furthermore, two or more terminal devices that collect characteristic information may be used. That is, two or more terminal devices may collect independent words and produce the independent word characteristic DB 32 that combines them.

  In the above method, a conversion formula that increases the score related to the preferred word is used. Conversely, a conversion formula that decreases the score related to the non-priority word (an independent word that is not the preferred word) may be used. Good. That is, the score may be adjusted so that the score related to the priority word is relatively larger than the score related to the non-priority word. Furthermore, in the above method, the score may be adjusted without determining whether or not it is a priority word. For example, the score may be adjusted according to the priority. Specifically, a multiple m and a constant a in the score conversion formula may be determined according to the priority value. In other words, the priority may be included in the score conversion formula, and the score of the related independent words with high priority may be relatively increased. Thus, the weight of the language model may be adjusted according to the priority value. Even in this case, the voice recognition performance can be improved for the same reason as described above.

Embodiment 2. FIG.
In the above embodiment, the correction coefficients j _{1 to} j _l are calculated based on the characteristic information stored in the independent word characteristic DB 32. In the present embodiment, based on the recognition result history 15, and calculates the different correction coefficients k ₁ to k _m. That is, using the correction coefficient _j 1 to j _l and the correction coefficient _k 1 to k _m, has a priority. This embodiment will be described with reference to FIGS. 7 and 8 are diagrams for explaining the speech recognition process and the priority calculation process. Note that a description of the same contents as those in the processing and configuration described in Embodiment 1 is omitted.

  As shown in FIG. 7, the speech recognition unit 10 calculates the likelihood of the independent word recognized in the speech recognition process. For example, among “I went to Okinawa”, the likelihood of “Okinawa” is 0.6, and “Go” is 0.5. The voice recognition unit 10 similarly determines the likelihood for other independent words, for example, “family”, “travel”, and the like.

  The recognition result history 15 stores the likelihood and the appearance time (occurrence time) in association with the independent word. The recognition result history 15 indicates that, for example, at 12:00:00, the independent word “Okinawa” was recognized with a likelihood of 0.6, and the independent word “go” was recognized with 0.5. I remember it. Similarly, the recognition result history 15 shows that at 12:06:20, the independent word “abdomen” was recognized with a likelihood of 0.5, and the independent word “suku” was recognized with a likelihood of 0.6. Is remembered. The recognition result history 15 similarly stores likelihood and appearance time corresponding to other independent words such as “family” and “travel”.

Priority correction unit 25 based on the correction information included in the recognition result history 15 calculates a correction coefficient k ₁ to k _m. Then, the priority Priority (w) is calculated based on the equation (2).

Specifically, the sum of the product of ten relevance and its correction coefficient k ₁ to k _m, is calculated. A value obtained by multiplying the sum by the correction coefficients j _{1 to} j _m is the priority. Similarly, priority is similarly calculated | required about all the independent words contained in independent word relevance DB31. The priority is a value indicating how much the related independent words stored in the independent word relevance DB 31 are related to the plurality of extracted independent words included in the recognition result history 15.

The following describes configuration examples of the correction coefficient _k 1 to k _m. The priority correction unit 25 determines the correction coefficient k ₁ based on the likelihood of the recognition result history 15. That is, the likelihood of an independent word stored in the recognition result history 15 is used as correction information. As shown in FIG. 7, the likelihood of “Soki soba” is 0.5, the likelihood of “eat” is 0.5, and the likelihood of “clean” is 0.6. And the priority correction | amendment part 25 calculates a correction coefficient with the comparison result of likelihood and a threshold value.

When the likelihood is 0.3 or more, k ₁ = 1.0 is defined, and when the likelihood is less than 0.3, k ₁ = 0.9 is defined. Thus, the correction coefficient k ₁ is set for each independent word based on the likelihood. In the above example, the threshold for likelihood is set to 0.3, but the threshold is not particularly limited. The correction coefficient may be set in multiple stages according to the comparison result with a plurality of threshold values. Further, the correction coefficient may be obtained by a function having the likelihood as a variable instead of the comparison result between the likelihood and the threshold.

Correction coefficient k ₂ is defined by the time of appearance of independent words. For example, k ₂ = 1.0 when the difference between the appearance time of the independent word and the current time is less than 10 seconds, and k ₂ = 0.8 when the difference is 10 seconds or more. In this way, the appearance time of an independent word can be used as correction information. An independent word whose appearance time is close to the current time increases the correction coefficient k ₂ , and an independent word far from the current time decreases the correction coefficient k ₂ . Of course, by setting a number of times that the threshold value may be set a correction coefficient k ₂ in multiple stages.

Furthermore, the appearance order of independent words can be used as correction information. For example, when the order of appearance in the recognition result history 15 is up to two immediately before, k ₃ = 1.0, and when it is three or more, k ₃ = 0.7 is defined. Thus, the priority correction unit 25 defines the appearance time of the independent words the correction factor k _3. For independent words whose appearance order is recent, the correction coefficient k ₃ is increased, and for independent words whose appearance order is old, the correction coefficient k ₃ is decreased. Of course, by setting a number of times that the threshold value may be set the correction coefficient k ₃ in multiple stages.

For example, in the above example, Priority (Shisar) = 1.6 is calculated. Since the threshold value Th is set to 1.0 in advance, “Shisar” is a priority word. The score is corrected in the same manner for this priority word. In the above example, the example using the correction coefficients k _{1 to} k ₃ has been described, but the number of correction coefficients is not limited to three. For example, at least one of the correction coefficients k _{1 to} k ₃ may be used, or two or more correction coefficients may be used in combination. Further, other correction information may be extracted from the recognition result history 15 and the correction coefficient may be used. Since the independent words highly related to the actual history of the speech recognition results are weighted, the speech recognition performance can be further improved. Even if the topic changes, it can respond appropriately. Furthermore, since the priority correction coefficient is calculated using the correction information acquired from the recognition result history 15, the speech recognition performance can be improved.

  Furthermore, the above-described voice recognition processing may be realized by causing a computer including a DSP (Digital Signal Processor), an MPU (Micro Processing Unit), a CPU (Central Processing Unit), or a combination thereof to execute a program.

  In the above example, a program including a group of instructions for causing a computer to perform speech recognition processing is stored using various types of non-transitory computer readable media and supplied to the computer. can do. Non-transitory computer readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (for example, flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (for example, magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-R / W, semiconductor memory (for example, mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM (Random Access Memory)) are included. The program may also be supplied to the computer by various types of transitory computer readable media. Examples of transitory computer readable media include electrical signals, optical signals, and electromagnetic waves. The temporary computer-readable medium can supply the program to the computer via a wired communication path such as an electric wire and an optical fiber, or a wireless communication path.

  The present invention is not limited to the above-described embodiment, and can be appropriately changed and combined without departing from the spirit of the present invention.

DESCRIPTION OF SYMBOLS 10 Speech recognition part 11 Feature-value extraction part 12 Similarity calculation part 13 Acoustic model 14 Ngram correction model 15 Recognition result log | history 16 Ngram language model 20 Preferred word estimation part 21 Independent word extraction part 22 Priority calculation part 23 Priority word discrimination | determination part 24 Language model update processing unit 25 Priority correction unit 30 Ontology 31 Independent word relevance DB
32 Independent word characteristics DB
40 Knowledge Learning Unit 41 User Request 42 Independent Word Extraction Unit 43 History Management Unit

Claims (9)

A speech recognition means for recognizing a user's speech using a language model;
Extraction means for extracting independent words contained in the user's voice;
Relevance storage means for storing independent words related to each other in association with the relevance;
Independent word characteristic storage means for storing characteristic information of independent words included in sentences used by the user;
Priority calculation means for calculating the priority of related independent words related to the independent words extracted by the extraction means using the characteristic information and the relevance;
A speech recognition apparatus comprising: adjusting means for adjusting a weight of the language model according to the priority.   The independent word characteristic storage means stores at least one of time information of the independent word, position information of the independent word, and user information of a user who uses the independent word as the characteristic information. The speech recognition apparatus according to claim 1. According to the priority of the related independent words, further comprising a determination means for determining whether the related independent words are priority words,
The speech recognition apparatus according to claim 1, wherein the adjustment unit updates the language model so as to relatively increase a score related to the priority word. Based on a comparison result between the characteristic information and current information, a correction coefficient is calculated,
For the independent words stored in the association degree storage unit, the priority is calculated by multiplying the sum of the degrees of association with the plurality of independent words extracted by the extraction unit by the correction coefficient,
The speech recognition apparatus according to claim 3, wherein the determination unit determines whether the related independent word is the priority word based on a comparison result between the priority and a threshold value. Based on a comparison result between the characteristic information and current information, a correction coefficient is calculated,
For the independent words stored in the association degree storage unit, the priority is calculated by multiplying the sum of the degrees of association with the plurality of independent words extracted by the extraction unit by the correction coefficient,
5. The voice according to claim 3, wherein the discriminating unit discriminates, from among the related independent words, the independent words having the highest priority N (N is a natural number) as the priority words. 6. Recognition device.   The speech recognition apparatus according to claim 1, wherein the independent words are limited to nouns, adjectives, and verbs.   The speech recognition apparatus according to claim 1, wherein the degree of association is set according to a co-occurrence frequency of two independent words in a sentence. Recognizing user speech using a language model;
Extracting independent words contained in the user's voice;
The degree of association between the related independent words stored in the related degree storage means and the characteristic information of the independent words included in the sentences used by the user stored in the independent word characteristic storage means, are extracted. Calculating priorities of related independent words associated with independent words;
Adjusting the weight of the language model according to the priority. Against the computer,
Recognizing user speech using a language model;
Extracting independent words contained in the user's voice;
The degree of association between the related independent words stored in the related degree storage means and the characteristic information of the independent words included in the sentences used by the user stored in the independent word characteristic storage means, are extracted. Calculating priorities of related independent words associated with independent words;
Adjusting the weight of the language model according to the priority;
Voice recognition program that executes

Images