WO2010061507A1 - 言語モデル作成装置 - Google Patents
言語モデル作成装置 Download PDFInfo
- Publication number
- WO2010061507A1 WO2010061507A1 PCT/JP2009/004341 JP2009004341W WO2010061507A1 WO 2010061507 A1 WO2010061507 A1 WO 2010061507A1 JP 2009004341 W JP2009004341 W JP 2009004341W WO 2010061507 A1 WO2010061507 A1 WO 2010061507A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- language model
- content
- word string
- probability
- speech recognition
- Prior art date
Links
- 238000000034 method Methods 0.000 claims description 71
- 230000008569 process Effects 0.000 claims description 58
- 238000012804 iterative process Methods 0.000 claims description 8
- 230000010365 information processing Effects 0.000 claims description 6
- 239000002344 surface layer Substances 0.000 claims description 4
- 238000005290 field theory Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 13
- 238000004364 calculation method Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 4
- 230000014509 gene expression Effects 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/18—Speech classification or search using natural language modelling
- G10L15/183—Speech classification or search using natural language modelling using context dependencies, e.g. language models
- G10L15/19—Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules
- G10L15/197—Probabilistic grammars, e.g. word n-grams
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/18—Speech classification or search using natural language modelling
- G10L15/1815—Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/18—Speech classification or search using natural language modelling
- G10L15/183—Speech classification or search using natural language modelling using context dependencies, e.g. language models
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/18—Speech classification or search using natural language modelling
- G10L15/183—Speech classification or search using natural language modelling using context dependencies, e.g. language models
- G10L15/19—Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules
Definitions
- the present invention relates to a language model creation device for creating a language model used for performing speech recognition processing for recognizing a word string corresponding to speech.
- a speech recognition device that recognizes a word string represented by speech from speech (utterance) uttered by a user is known.
- the speech recognition device described in Patent Document 1 as one of this type of speech recognition device performs speech recognition processing for recognizing a word string corresponding to speech based on a plurality of pre-stored language models. .
- the content-specific language model is a model that represents the probability of occurrence of a specific word in a word string that represents specific content (topic, keyword, etc.). For example, in a word sequence that includes a TV program, the probability that a program name or a talent name will appear is high, and in a word sequence that includes sports, a team name, an exercise equipment name, or a player name is included. The probability of appearing is increased.
- the content may change in a series of voices uttered by the user.
- speech recognition processing is performed based only on one content-specific language model, the accuracy of recognizing word strings may be excessively reduced.
- the speech recognition apparatus is configured to use a language model for different contents for each predetermined section in one utterance.
- the speech recognition apparatus in order to determine which content-specific language model is used, a process of evaluating a recognition result when each content-specific language model is used is performed. Therefore, the speech recognition apparatus has a problem that the processing load for determining which content-specific language model is used is excessive.
- the object of the present invention is to solve the above-mentioned problem “the calculation load for creating a language model becomes excessive and the word string cannot be recognized from speech with high accuracy”. It is an object of the present invention to provide a language model creation device that can do this.
- a language model creation device includes: A first content-specific language model that represents the probability that a specific word appears in a word string that represents the first content, and a second content that represents the probability that the specific word appears in a word string that represents the second content A different language model, a content-specific language model storage means for storing, A processing target word string that is at least part of an input word string that is a word string included in a speech recognition hypothesis generated by performing a voice recognition process that recognizes a word string corresponding to speech A first probability parameter representing the probability that the content represented by the first content is the first probability parameter, and a second probability parameter representing the probability that the content represented by the processing target word string is the second content.
- the acquired first probability parameter, the acquired second probability parameter, the stored first content-specific language model, and the stored second content-specific language model And performing language model creation processing for creating a language model representing the probability that the specific word appears in a word string corresponding to a portion corresponding to the processing target word string in the speech.
- a language model creation means Is provided.
- a speech recognition apparatus A first content-specific language model that represents the probability that a specific word appears in a word string that represents the first content, and a second content that represents the probability that the specific word appears in a word string that represents the second content
- a different language model a content-specific language model storage means for storing,
- a processing target word string that is at least part of an input word string that is a word string included in a speech recognition hypothesis generated by performing a voice recognition process that recognizes a word string corresponding to speech
- a first probability parameter representing the probability that the content represented by the first content is the first probability parameter
- a second probability parameter representing the probability that the content represented by the processing target word string is the second content.
- the acquired first probability parameter, the acquired second probability parameter, the stored first content-specific language model, and the stored second content-specific language model And performing language model creation processing for creating a language model representing the probability that the specific word appears in a word string corresponding to a portion corresponding to the processing target word string in the speech.
- a language model creation means Speech recognition means for performing speech recognition processing for recognizing a word string corresponding to the input speech based on the language model created by the language model creation means; Is provided.
- a language model creation method includes: A first content-specific language model that represents the probability that a specific word appears in a word string that represents the first content, and a second content that represents the probability that the specific word appears in a word string that represents the second content
- a processing target word string that is at least part of an input word string that is a word string included in a speech recognition hypothesis generated by performing a voice recognition process that recognizes a word string corresponding to speech
- a first probability parameter representing the probability that the content represented by the first content is the first probability parameter
- a second probability parameter representing the probability that the content represented by the processing target word string is the second content.
- the acquired first probability parameter, the acquired second probability parameter, the stored first content-specific language model, and the stored second content-specific language model And a language model representing the probability that the specific word appears in a word string corresponding to a portion corresponding to the processing target word string in the speech.
- a language model creation program includes: In the information processing device, A first content-specific language model that represents the probability that a specific word appears in a word string that represents the first content, and a second content that represents the probability that the specific word appears in a word string that represents the second content A different language model, and a content-specific language model storage processing unit for storing the different language model in a storage device; A processing target word string that is at least part of an input word string that is a word string included in a speech recognition hypothesis generated by performing a voice recognition process that recognizes a word string corresponding to speech A first probability parameter representing the probability that the content represented by the first content is the first probability parameter, and a second probability parameter representing the probability that the content represented by the processing target word string is the second content.
- the acquired first probability parameter, the acquired second probability parameter, the stored first content-specific language model, and the stored second content-specific language model And a language model creating means for creating a language model representing a probability that the specific word appears in a word string corresponding to a portion corresponding to the processing target word string in the speech, It is a program for realizing.
- the present invention can create a language model that can recognize a word string corresponding to speech with high accuracy while preventing an excessive calculation load. it can.
- FIG. 1 It is a block diagram showing the outline of the function of the language model creation apparatus which concerns on 1st Embodiment of this invention. It is the flowchart which showed the action
- the language model creation device 101 is an information processing device.
- the language model creating apparatus 101 includes a central processing unit (CPU) (not shown), a storage device (memory and hard disk drive (HDD)), an input device, and an output device.
- CPU central processing unit
- HDD hard disk drive
- the output device has a display.
- the output device displays an image made up of characters and graphics on the display based on the image information output by the CPU.
- the input device has a mouse, keyboard and microphone.
- the language model creation apparatus 101 is configured such that information based on a user operation is input via a keyboard and a mouse.
- the language model creation device 101 is configured such that input speech information representing speech around the microphone (that is, outside the language model creation device 101) is input via the microphone.
- the language model creation apparatus 101 accepts a speech recognition hypothesis (input word string) including a word string as a speech recognition result output by a speech recognition apparatus (not shown), and responds to the accepted speech recognition hypothesis.
- the language model is output to the speech recognition apparatus.
- the function of the language model creation device 101 includes a speech recognition hypothesis input unit (part of a language model creation unit) 11, a content estimation unit (part of a language model creation unit) 12, Language model creation unit (part of language model creation means) 13, content model storage unit (part of language model creation means) 14, and content-specific language model storage unit (content-specific language model storage means, content-specific language model Storage processing means, content-specific language model storage processing step) 15.
- This function is realized by the CPU of the language model creation device 101 executing a program stored in the storage device.
- This function may be realized by hardware such as a logic circuit.
- the speech recognition hypothesis input unit 11 receives a speech recognition hypothesis (input word sequence) including a word string as a speech recognition result output by an external speech recognition device (not shown), and estimates the content of the received speech recognition hypothesis To the unit 12.
- the speech recognition hypothesis is information generated by performing speech recognition processing in which the speech recognition apparatus recognizes a word string corresponding to speech.
- the speech recognition hypothesis is information representing a word string composed of one or more words.
- the speech recognition hypothesis may be information representing a plurality of word strings (for example, a word graph or N best word strings (best N word strings)).
- the content estimation unit 12 divides the speech recognition hypothesis output from the speech recognition hypothesis input unit 11 with the boundary between words of the word sequence recognized by the speech recognition process as a boundary, thereby obtaining at least one processing target word sequence. Extract (generate) from speech recognition hypothesis. According to this, when creating a language model, the information acquired when performing speech recognition processing can be used. As a result, the contents can be estimated with high accuracy, and a language model can be created quickly.
- the content estimation unit 12 extracts at least one processing target word string from the speech recognition hypothesis by dividing the speech recognition hypothesis using a position different from the boundary between words of the word string recognized by the speech recognition process as a boundary. (Generation) may be performed. According to this, even when the boundary between the actual words in the utterance is different from the boundary between the words in the word string recognized by the speech recognition processing, the word string corresponding to the speech is recognized with high accuracy. You can create a language model that makes it possible.
- the content estimation unit 12 has a probability that the content represented by the processing target word string is specific content (first content, second content, or the like) for each of the plurality of processing target word strings generated. Is calculated (acquired) based on the content model stored in the content model storage unit 14 (a first probability parameter, a second probability parameter, or the like). For example, the content estimation unit 12 calculates a first probability parameter that represents the probability that the content represented by the processing target word string is the first content, and the second probability parameter that represents the probability that the content is the second content. Is calculated. Then, the content estimation unit 12 outputs the acquired probability parameter to the language model creation unit 13.
- the probability parameter is a value of the probability that the content represented by the processing target word string is specific content.
- the probability parameter may be a value that increases as the probability that the content represented by the processing target word string is specific content increases. That is, it can be said that the probability parameter represents the likelihood that the content represented by the processing target word string is specific content.
- the probability parameter may be referred to as a likelihood parameter or a weight parameter.
- the contents are also called topics.
- the contents used as search conditions for searching for a television program include: a person name (talent name and group name, etc.), a program name, a program genre name (variety and sports, etc.), a broadcast station name, And time expressions (evening and 8 o'clock, etc.), etc. If the contents are different, the probability that a specific word string appears (exists) during utterance is different.
- the content estimation unit 12 estimates the probability that the content represented by the word string in the section is specific content for each section (processing target word string) during the utterance. Therefore, even when the content changes during the utterance, the probability can be estimated with high accuracy for each section.
- the content model storage unit 14 stores in advance a content model (information) representing a relationship between a word string and the probability that the content represented by the word string is each of a plurality of contents.
- the content model is a probability model based on the theory of conditional random fields (CRF).
- CRF conditional random fields
- X is a processing target word string
- Y is the content. That is, the right side P (Y
- ⁇ (X, Y) is information representing the feature (feature) of the processing target word string X
- ⁇ is a model parameter (weight value) in the CRF corresponding to each of the features ⁇ (X, Y).
- Z is a normalization term. Note that “exp ()” indicates a function for obtaining the power of a numerical value with e as the base.
- the content model storage unit 14 stores the feature ⁇ and the model parameter ⁇ (weight value) in the storage device.
- the content estimation unit 12 estimates the contents represented by (belonging to) each word of the speech recognition hypothesis when the speech recognition hypothesis is a word string and CRF is used as the content model.
- the content estimation unit 12 first develops a section corresponding to each word included in the word string of the speech recognition hypothesis into candidate content and holds it in a graph format (content candidate graph).
- FIG. 3 is an example of a word string of a speech recognition hypothesis
- FIG. 4 is an example of a content candidate graph.
- FIG. 3 shows a part of the word string of the speech recognition hypothesis.
- the content estimation unit 12 generates three types of “person name”, “program name”, and “others” as content candidates by expanding each segment.
- the arc (arc, edge) A in FIG. 4 indicates that the content represented by the word “country travel” in the speech recognition hypothesis is “person name” as the content.
- the content estimation unit 12 ranks and outputs the content path (content sequence) represented by the content candidate graph based on a predetermined criterion (for example, a score calculated with reference to the content model). Specifically, the content estimation unit 12 obtains a score with reference to the content model in each arc in the graph, and accumulates the score for each path.
- a predetermined criterion for example, a score calculated with reference to the content model.
- the content estimation unit 12 specifies a path that maximizes the left side P (Y
- the score at each arc in the content candidate graph is the product of the feature (feature) related to each arc and the weight value for each feature that is a model parameter of the CRF.
- FIG. 5 is an example of the characteristics related to the arc A.
- These features are expressed as CRF features ( ⁇ ) as shown in FIG.
- the score of the arc A is calculated by the product of the value of these features and the weight ⁇ of the “person name” corresponding to the arc A among the model parameters. The larger the score, the higher the content likelihood.
- the content estimation unit 12 acquires the probability parameter based on the word attribute information indicating the attribute of the word included in the processing target word string.
- the word attribute information includes at least one of information representing the word surface layer, information representing the reading, and information representing the part of speech.
- the content estimation unit 12 may acquire a probability parameter based on the speech recognition processing information acquired when performing speech recognition processing for recognizing a processing target word string from speech.
- the speech recognition processing information includes information representing the recognition reliability that is the reliability of the recognition result by the speech recognition processing, information representing the duration of time for which one sound continues, and presence or absence of preceding silence At least one piece of information representing
- the target section not only local features related to the target section, but also global features related to the entire speech recognition hypothesis (the entire utterance), positional information within the speech recognition hypothesis (first half, second half, etc.), co-occurrence within the utterance Word information, information related to the structure of the word graph (average number of branches of arcs, etc.), content connection information, and the like may be used as the feature ( ⁇ ) of the content model.
- the content estimation unit 12 is information indicating the position of the processing target word string in the input word string, information indicating that the input word string includes a plurality of words, information indicating the connection state of the contents in the input word string, And a probability parameter may be acquired based on at least one of the information showing the relationship between each input word sequence when there are a plurality of input word sequences.
- the a posteriori appearance probability (a posteriori probability) p (Yi c
- X) of each arc of the content candidate graph is calculated by recursive calculation using the Forward algorithm and the Backward algorithm.
- the content estimation unit 12 uses this probability p as the appearance probability ( Used as a probability parameter).
- the model parameter of CRF is iteratively calculated according to the criterion for maximizing the log likelihood of the above equation (1) using a pair of input (X: word string) and output (Y: content) associated in advance as learning data. It may be optimized (learned) by a method or the like.
- the language model creation unit 13 includes a content estimation result including the probability parameters (for example, the first probability parameter and the second probability parameter) output from the content estimation unit 12, and the content-specific language model storage unit 15 Represents the probability that a specific word will appear in a word sequence corresponding to a portion corresponding to the processing target word sequence in the speech that is the basis of the input word sequence
- a language model creation process for creating a language model is performed for each word string to be processed. Then, the language model creation unit 13 outputs the created language model.
- the content-specific language model and the language model are N-gram language models that are constructed based on the assumption that the probability that a certain word appears depends only on the immediately preceding N ⁇ 1 words. It is.
- the appearance probability of the i-th word w i is represented by P (w i
- W i ⁇ N + 1 i ⁇ 1 in the condition part represents the (i ⁇ N + 1) to (i ⁇ 1) th word string.
- a model constructed based on the assumption that it is not affected by the immediately preceding word is called a unigram model.
- the parameters made up of various conditional probabilities of various words used in the N-gram language model are obtained by maximum likelihood estimation for the learning text data.
- the content-specific language model storage unit 15 stores a plurality of content-specific language models in a storage device in advance.
- the plurality of content-specific language models are models representing the probability that a specific word appears in a word string representing different content. That is, the plurality of content-specific language models include a first content-specific language model that represents a probability that a specific word appears in a word string that represents the first content, and a specific word that represents the second content. And a second language model for content representing the probability of occurrence of.
- each content language model is a trigram model.
- the language model creation unit 13 stores a score for each content in each section (that is, a probability parameter indicating the probability that the content represented by each processing target word string is each content) and the content-specific language model storage unit 15.
- a language model is created according to the following equation (3) from the plurality of language models by content.
- Equation (3) P t (w i ) is the probability that the word w i will appear in the section “t”, and ⁇ j (t) is the content “j” that the word string in the section “t” represents.
- P j (w i ) is the probability of appearance of word w i in the content-specific language model for content “j”.
- the language model creation unit 13 uses the probability parameter acquired by the content estimation unit 12 (the content appearance probability in each section (processing target word string) in the utterance) as ⁇ j (t) in Expression (3). ).
- the language model creating unit 13 has a probability that the first content-specific language model represents the first coefficient (for example, the first probability parameter) that increases as the calculated first probability parameter increases.
- the value multiplied by P j (w i )) in equation (3) and the second coefficient (for example, the second probability parameter) that increases as the calculated second probability parameter increases are classified according to the second contents.
- the probability that a specific word appears in the word sequence corresponding to the portion corresponding to the processing target word sequence in the speech that is the basis of the input word sequence, as the sum of the value multiplied by the probability represented by the language model increases. Create a language model that increases
- t in the expression (3) may be a section corresponding to a time frame used in the speech recognition processing, or may be a time representing a time point in the utterance.
- the content-specific language model storage unit 15 may store a content-specific language model and a list (word list) of words having a high probability of appearing for each content.
- the language model creation unit 13 may be configured to increase the probability that a word included in the word list for the content having the highest score appears in each section in the utterance by a predetermined value.
- the content estimation unit 12 may change the value estimated for each section as the above-described score (content appearance score) without using it as it is.
- a word string in a certain utterance may include a word string that is not related to any content.
- the content estimation unit 12 estimates the content represented by the word string from a total of four types, that is, three types of content and “none of the content”. Then, the content estimation unit 12 uses a predetermined value (for example, a predetermined ratio) as the score of the other three types of content in the section in which the content represented by the word string is estimated to be “no content”. (For example, a value according to a certain ratio) may be changed.
- the content estimation unit 12 may set all of the calculated probability parameters to the same value when all of the calculated probability parameters (scores) are smaller than a preset lower limit threshold.
- the content estimation unit 12 sets all the probability parameters for the section t2 to the same value (in this example, the lower limit threshold value).
- the content represented by the word string is the content “person name” related to the search condition of the television program
- a word such as “appearing” or “out” follows the word string.
- the chance of appearing is relatively high. Therefore, in the section following the section in which the content represented by the word string is estimated to be the content “person name”, it is desirable that the “person name” score immediately decreases in order to recognize the subsequent word string with high accuracy. Absent.
- the content estimation unit 12 is adjacent to the processing target word string. You may correct
- the content estimation unit 12 corrects the score acquired for the section t1 and the section t3 adjacent to the section t2 to increase.
- the content estimation unit 12 corrects the score so that the closer to the section t2 in the section t1, the closer the score is to the score acquired for the section t2. Similarly, in the section t3, the content estimation unit 12 corrects the score so that the closer to the section t2, the closer the score is to the score acquired for the section t2.
- a language model can be created.
- the language model creation unit 13 may output all the information included in the language model when outputting the created language model, or may output only information specified from the outside.
- the language model creation apparatus 101 when activated, the language model creation apparatus 101 reads the content model and the content-specific language model from the storage device that implements the content model storage unit 14 and the content-specific language model storage unit 15. , Initialization processing is performed for referring to each from the content estimation unit 12 and the language model creation unit 13 (step S11).
- the speech recognition hypothesis input unit 11 accepts a speech recognition hypothesis from an external speech recognition device in response to a notification indicating the end of speech recognition processing, and outputs the accepted speech recognition hypothesis to the content estimation unit 12 (step S12). Part of the language model creation process). Note that the speech recognition hypothesis input unit 11 may be configured to accept a speech recognition hypothesis input by the user.
- the content estimation unit 12 selects each processing target word string in the speech recognition hypothesis based on the content model stored in the content model storage unit 14.
- a probability parameter representing the probability that the content represented by (for example, each word) is a specific content is calculated (step S13, part of the language model creation step).
- the language model creation unit 13 determines the basis of the speech recognition hypothesis based on the probability parameter output from the content estimation unit 12 and the content-specific language model stored in the content-specific language model storage unit 15.
- a language model representing the probability that a specific word appears in the word string corresponding to the portion corresponding to the processing target word string in the generated speech is generated, and the created language model is output (step S14, language model creation step) Part of).
- the language model creation device 101 includes the probability that the content represented by the processing target word string is the first content, and the processing target word.
- a language model is created based on the probability that the content represented by the column is the second content, the first content-specific language model, and the second content-specific language model.
- the language model creation device 101 processing load can be reduced.
- the language model creating apparatus 101 it is possible to create a language model that can recognize a word string corresponding to speech with high accuracy while preventing an excessive calculation load.
- the probability that the content represented by the processing target word string is the first content increases, the probability that the probability represented by the first language-specific language model is reflected in the language model increases. can do.
- the greater the probability that the content represented by the processing target word string is the second content the greater the degree that the probability represented by the second content-specific language model is reflected in the language model.
- FIG. 9 is a block diagram showing functions of the speech recognition apparatus 201 according to the second embodiment of the present invention.
- the speech recognition apparatus 201 is an information processing apparatus having the same configuration as the language model creation apparatus 101 according to the first embodiment.
- the functions of the speech recognition apparatus 201 include a speech recognition unit (speech recognition unit) 21, a speech recognition model storage unit 22, and a language model update unit (language model creation unit) 24.
- the speech recognition apparatus 201 generates a speech recognition hypothesis as an input word sequence by performing speech recognition processing for recognizing a word sequence corresponding to the input speech, and based on the generated speech recognition hypothesis, the first recognition Similar to the language model creation apparatus 101 according to the embodiment, a language model is created, and the speech recognition process is performed again based on the created language model.
- the voice recognition unit 21 generates an input word string as a voice recognition hypothesis (for example, a word graph) by performing the voice recognition process for recognizing a word string corresponding to the voice input via the input device.
- the voice recognition unit 21 may be configured to receive voice by receiving voice information representing voice from another information processing apparatus.
- the speech recognition unit 21 is a model stored in the speech recognition model storage unit 22 (a model for performing speech recognition processing, including an acoustic model, a language model, a word dictionary, etc.) for all sections of speech.
- the speech recognition process is performed by searching for a word string that matches the speech according to the score calculated based on (1).
- the acoustic model is a hidden Markov model
- the language model is a word trigram.
- the speech recognition unit 21 refers to the language model output by the language model update unit 24 when performing speech recognition processing. For example, when the speech recognition unit 21 calculates the probability of occurrence of the word w i with reference to the language model of the formula (3) in a certain time frame “f” during speech recognition processing, the speech recognition unit 21 corresponds to the “f”. P t (w i ) is referred to for the section “t” to be performed. In this example, the time frame represents a unit for converting the speech to be recognized into a feature value for recognition.
- the speech recognition unit 21 refers to the language model stored in the speech recognition model storage unit 22 when performing speech recognition processing before the language model update unit 24 creates a language model corresponding to the utterance.
- the speech recognition unit 21 may be configured to use the sum of the probabilities represented by the plurality of content-specific language models stored in the content-specific language model storage unit 245 as the probability that a word will appear.
- the speech recognition device 201 Based on the language model created by the language model update unit 24, the speech recognition device 201 recognizes the speech recognition processing by the speech recognition unit 21 and the speech recognition unit 21 that recognizes a word string corresponding to the input speech. On the basis of the word string, the language model update unit 24 executes an iterative process that alternately repeats a language model creation process for creating a language model.
- the first probability parameter and the second probability parameter can be acquired with higher accuracy as the accuracy of the input word sequence (the degree of matching with the true word sequence) increases. Furthermore, a language model that makes it possible to recognize a word string corresponding to speech with higher accuracy as the accuracy of the first probability parameter and the second probability parameter becomes higher. Therefore, according to the above configuration, it is possible to recognize a word string corresponding to speech with higher accuracy.
- the speech recognition unit 21 terminates the iterative process when a predetermined termination condition is established based on the generated speech recognition hypothesis or the language model used in the speech recognition process. When completing the iterative process, the speech recognition unit 21 outputs the latest speech recognition hypothesis acquired at that time as a speech recognition result. The speech recognition unit 21 may select and output a speech recognition result from speech recognition hypotheses accumulated up to that point.
- the end condition is a condition that the word string recognized by the previous voice recognition process matches the word string recognized by the current voice recognition process.
- the end condition may be a condition that the number of times the voice recognition process is executed is larger than a preset threshold number.
- the termination condition may be a condition determined based on a language model parameter created by the language model creation unit 243, an estimation result output by the content estimation unit 242, or a score thereof.
- the language model update unit 24 has the same function as the language model creation apparatus 101 according to the first embodiment.
- the language model update unit 24 includes a speech recognition hypothesis input unit 241 similar to the speech recognition hypothesis input unit 11, a content estimation unit 242 similar to the content estimation unit 12, and a language model creation unit 243 similar to the language model creation unit 13.
- the language model update unit 24 stores the speech recognition hypothesis output from the speech recognition unit 21, the stored content model, and the stored content A language model is created based on the language model, and the created language model is output.
- the content-specific language model storage unit 245 stores a word trigram in the storage device as a content-specific language model.
- the language model creation unit 243 has a score indicating the probability that the content represented by the processing target word string is a specific content, the stored language model for each content, the above formula (3), Create a language model based on
- the language model update unit 24 creates a language model based on the received speech recognition hypothesis each time a speech recognition hypothesis is received from the speech recognition unit 21 until the speech recognition unit 21 determines that the termination condition is satisfied. .
- the probability that the word w i appears is P t, k (w i ) (see the following equation (4)).
- the speech recognition unit 21 performs (k + 1) th speech recognition processing with reference to this language model, and outputs a speech recognition hypothesis.
- the content estimation unit 242 inputs this speech recognition hypothesis and outputs the appearance score ⁇ j, k + 1 (t) of each content as the (k + 1) th content estimation result.
- the language model creation unit 243 creates the (k + 1) -th language model P t, k + 1 (w i ) using this appearance score (see the following formula (5)). In this way, the accuracy of each is gradually increased by repeatedly updating the speech recognition hypothesis and the content estimation result.
- the speech recognition unit 21 when the speech recognition unit 21 performs the second and subsequent processing, the speech recognition unit 21 replaces the speech recognition processing with speech as an input, and the previous speech recognition hypothesis (word graph or the like). A re-score process may be performed.
- the speech recognition apparatus 201 when activated, obtains a speech recognition model and a language model from the speech recognition model storage unit 22 and the storage device that implements the content-specific language model storage unit 245. Read-out and initialization processing for referring to each from the speech recognition unit 21 and the language model update unit 24 are performed (step S21).
- the voice recognition unit 21 accepts voice input from the outside via the input device in response to a notification indicating the end of voice input (step S22).
- the voice recognition unit 21 When receiving the voice, the voice recognition unit 21 receives the received voice based on the voice recognition model stored in the voice recognition model storage unit 22 and the language model created by the language model update unit 24. Voice recognition processing is performed (step S23).
- the speech recognition apparatus 201 determines whether or not an end condition is satisfied based on the speech recognition hypothesis output by the speech recognition unit 21 performing speech recognition processing (step S24). If the termination condition is satisfied, the speech recognition apparatus 201 determines “Yes” and outputs the latest speech recognition hypothesis acquired at that time as a speech recognition result (step S27).
- the speech recognition apparatus 201 determines “No” in step S24 and creates a language model by executing the processes of steps S25 and S26. This process is the same as the process of steps S13 and S14 of FIG.
- the speech recognition apparatus 201 has the probability that the content represented by the processing target word string is the first content and the processing target word string.
- a language model is created based on the probability that the content to be represented is the second content, the first language model by content, and the second language model by content.
- the speech recognition apparatus 201 performs speech recognition processing for recognizing a word string corresponding to speech based on the created language model. Thereby, it is possible to recognize the word string corresponding to the voice with high accuracy while preventing the calculation load of the voice recognition device 201 from becoming excessive.
- the probability that the content represented by the processing target word string is the first content increases, the probability that the probability represented by the first language-specific language model is reflected in the language model increases. can do.
- the greater the probability that the content represented by the processing target word string is the second content the greater the degree that the probability represented by the second content-specific language model is reflected in the language model.
- the speech recognition apparatus 201 includes a speech recognition process for recognizing a word string corresponding to the speech input by the speech recognition unit 21 based on the language model created by the language model update unit 24, and the speech recognition unit 21.
- the language model update unit 24 executes an iterative process that alternately repeats a language model creation process for creating a language model.
- the higher the accuracy of the input word string (the degree of matching with the true word string), the higher the accuracy of the first probability parameter and the second probability parameter can be obtained. Furthermore, a language model that makes it possible to recognize a word string corresponding to speech with higher accuracy as the accuracy of the first probability parameter and the second probability parameter becomes higher. Therefore, according to the above configuration, it is possible to recognize a word string corresponding to speech with higher accuracy.
- the functions of the language model creation device 301 according to the third embodiment include a content-specific language model storage unit (content-specific language model storage unit) 35 and a language model creation unit (language model creation unit) 33.
- the content-specific language model storage unit 35 includes a first content-specific language model that represents a probability that a specific word appears in a word string that represents the first content, and a word string that represents the second content.
- a second content-specific language model representing the probability of appearance is stored in the storage device.
- the language model creation unit 33 includes at least one input word string that is a word string included in the speech recognition hypothesis generated by performing a speech recognition process for recognizing a word string corresponding to speech and is an input word string.
- a first probability parameter representing the probability that the content represented by the processing target word string is the first content
- a second probability parameter representing the probability that the content represented by the processing target word string is the second content And get.
- the language model creation unit 33 includes the acquired first probability parameter, the acquired second probability parameter, the first content-specific language model stored in the content-specific language model storage unit 35, and the content Based on the second content-specific language model stored in the different language model storage unit 35, a specific word appears in a word string corresponding to a portion corresponding to the processing target word string in the speech.
- a language model creation process for creating a language model representing a probability is performed.
- the language model creation device 301 has a probability that the content represented by the processing target word string is the first content, a probability that the content represented by the processing target word string is the second content, and the first content.
- a language model is created based on the different language model and the second content-specific language model.
- the language model creation device Since it is not necessary to perform a process of evaluating a recognition result when using each content-specific language model in order to determine which content-specific language model is used, the language model creation device The processing load 301 can be reduced.
- the language model creation device 301 it is possible to create a language model that can recognize a word string corresponding to speech with high accuracy while preventing an excessive calculation load.
- the language model creating means is The value obtained by multiplying the probability represented by the first language model by contents by the first coefficient, which increases as the acquired first probability parameter increases, and increases as the acquired second probability parameter increases.
- the above-mentioned specification is performed in the word string corresponding to the portion corresponding to the processing target word string in the speech. It is preferable that the language model is generated so that the probability of occurrence of the word is increased.
- the greater the probability that the content represented by the processing target word string is the first content the greater the degree to which the probability represented by the first content-specific language model is reflected in the language model.
- the greater the probability that the content represented by the processing target word string is the second content the greater the degree that the probability represented by the second content-specific language model is reflected in the language model.
- the language model creating means is configured to output the language for each of the plurality of processing target word strings obtained by dividing the input word string with a boundary between words of the word string recognized by the speech recognition process as a boundary. It is preferable to be configured to perform a model creation process.
- the language model creating means is configured to generate the language for each of the plurality of processing target word strings obtained by dividing the input word string with a position different from the boundary between words of the word string recognized by the speech recognition process as a boundary. It is preferable to be configured to perform a model creation process.
- the language model creating means is When both the acquired first probability parameter and the acquired second probability parameter are smaller than a preset lower threshold, the acquired first probability parameter and the acquired second Preferably, both probability parameters are set to the same value.
- the language model creating means is When the acquired first probability parameter for a certain processing target word string is larger than a preset upper threshold, the acquired first probability parameter is set for the processing target word string adjacent to the processing target word string. Preferably, the correction is made so that the probability parameter of 1 is increased.
- the word sequence corresponding to the speech is recognized with high accuracy even for the speech section corresponding to the processing target word sequence adjacent to the processing target word sequence identified with relatively high accuracy.
- the language model creating means is configured to acquire the first probability parameter and the second probability parameter based on a conditional random field theory.
- the language model creating means includes word attribute information representing the attributes of words included in the processing target word string, and voice acquired when performing voice recognition processing for recognizing the processing target word string from voice. It is preferable that the first probability parameter and the second probability parameter are acquired based on at least one of recognition processing information.
- the word attribute information preferably includes at least one of information representing the word surface layer, information representing the reading, and information representing the part of speech.
- the speech recognition processing information includes information representing the recognition reliability that is the reliability of the recognition result obtained by the speech recognition processing, information representing the duration of time for which one sound continues, and preceding silence. It is preferable to include at least one piece of information indicating the presence or absence of.
- the language model creating means is Information indicating the position of the processing target word string in the input word string, information indicating that the input word string includes a plurality of words, information indicating a connected state of contents in the input word string, and the input word It is preferable that the first probability parameter and the second probability parameter are acquired based on at least one of the information representing the relationship between the input word strings when there are a plurality of strings. is there.
- a speech recognition apparatus A first content-specific language model that represents the probability that a specific word appears in a word string that represents the first content, and a second content that represents the probability that the specific word appears in a word string that represents the second content
- a different language model a content-specific language model storage means for storing,
- a processing target word string that is at least part of an input word string that is a word string included in a speech recognition hypothesis generated by performing a voice recognition process that recognizes a word string corresponding to speech
- a first probability parameter representing the probability that the content represented by the first content is the first probability parameter
- a second probability parameter representing the probability that the content represented by the processing target word string is the second content.
- the acquired first probability parameter, the acquired second probability parameter, the stored first content-specific language model, and the stored second content-specific language model And performing language model creation processing for creating a language model representing the probability that the specific word appears in a word string corresponding to a portion corresponding to the processing target word string in the speech.
- a language model creation means Speech recognition means for performing speech recognition processing for recognizing a word string corresponding to the input speech based on the language model created by the language model creation means; Is provided.
- the speech recognition apparatus has a probability that the content represented by the processing target word string is the first content, a probability that the content represented by the processing target word string is the second content, and the first language by content.
- a language model is created based on the model and the second content-specific language model. Then, the speech recognition apparatus performs speech recognition processing for recognizing a word string corresponding to speech based on the created language model. Thereby, it is possible to recognize the word string corresponding to the voice with high accuracy while preventing the calculation load of the voice recognition device from becoming excessive.
- the language model creating means is The value obtained by multiplying the probability represented by the first language model by contents by the first coefficient, which increases as the acquired first probability parameter increases, and increases as the acquired second probability parameter increases.
- the above-mentioned specification is performed in the word string corresponding to the portion corresponding to the processing target word string in the speech. It is preferable that the language model is generated so that the probability of occurrence of the word is increased.
- the greater the probability that the content represented by the processing target word string is the first content the greater the degree to which the probability represented by the first content-specific language model is reflected in the language model.
- the greater the probability that the content represented by the processing target word string is the second content the greater the degree that the probability represented by the second content-specific language model is reflected in the language model.
- the voice recognition means is configured to generate the input word string by performing the voice recognition process for recognizing a word string corresponding to the input voice
- the language model creating means is configured to create the language model based on the input word string generated by the speech recognition means
- the voice recognition unit is configured to perform the voice recognition process for recognizing a word string corresponding to the input voice based on the language model created by the language model creation unit. It is.
- the voice recognition device Based on the language model created by the language model creating means, the speech recognition means recognizes the word string corresponding to the inputted speech, and the word string recognized by the speech recognition means Based on this, it is preferable that the language model creating means is configured to execute an iterative process that alternately repeats the language model creating process for creating the language model.
- the first probability parameter and the second probability parameter can be acquired with higher accuracy as the accuracy of the input word sequence (the degree of matching with the true word sequence) increases. Furthermore, a language model that makes it possible to recognize a word string corresponding to speech with higher accuracy as the accuracy of the first probability parameter and the second probability parameter becomes higher. Therefore, according to the above configuration, it is possible to recognize a word string corresponding to speech with higher accuracy.
- the voice recognition device is configured to end the iterative process when a predetermined end condition is satisfied.
- the end condition is a condition that the word string recognized by the previous voice recognition process matches the word string recognized by the current voice recognition process.
- the end condition is preferably a condition that the number of times the voice recognition process is executed is larger than a preset threshold number.
- a language model creation method includes: A first content-specific language model that represents the probability that a specific word appears in a word string that represents the first content, and a second content that represents the probability that the specific word appears in a word string that represents the second content
- a processing target word string that is at least part of an input word string that is a word string included in a speech recognition hypothesis generated by performing a voice recognition process that recognizes a word string corresponding to speech
- a first probability parameter representing the probability that the content represented by the first content is the first probability parameter
- a second probability parameter representing the probability that the content represented by the processing target word string is the second content.
- the acquired first probability parameter, the acquired second probability parameter, the stored first content-specific language model, and the stored second content-specific language model And a language model representing the probability that the specific word appears in a word string corresponding to a portion corresponding to the processing target word string in the speech.
- the language model creation method is The value obtained by multiplying the probability represented by the first language model by contents by the first coefficient, which increases as the acquired first probability parameter increases, and increases as the acquired second probability parameter increases.
- the above-mentioned specification is performed in the word string corresponding to the portion corresponding to the processing target word string in the speech. It is preferable that the language model is generated so that the probability of occurrence of the word is increased.
- a language model creation program includes: In the information processing device, A first content-specific language model that represents the probability that a specific word appears in a word string that represents the first content, and a second content that represents the probability that the specific word appears in a word string that represents the second content A different language model, and a content-specific language model storage processing unit for storing the same in a storage device; A processing target word string that is at least part of an input word string that is a word string included in a speech recognition hypothesis generated by performing a voice recognition process that recognizes a word string corresponding to speech A first probability parameter representing the probability that the content represented by the first content is the first probability parameter, and a second probability parameter representing the probability that the content represented by the processing target word string is the second content.
- the acquired first probability parameter, the acquired second probability parameter, the stored first content-specific language model, and the stored second content-specific language model And a language model creating means for creating a language model representing a probability that the specific word appears in a word string corresponding to a portion corresponding to the processing target word string in the speech, It is a program for realizing.
- the language model creating means is The value obtained by multiplying the probability represented by the first language model by contents by the first coefficient, which increases as the acquired first probability parameter increases, and increases as the acquired second probability parameter increases.
- the above-mentioned specification is performed in the word string corresponding to the portion corresponding to the processing target word string in the speech. It is preferable that the language model is generated so that the probability of occurrence of the word is increased.
- the program is stored in the storage device, but may be stored in a CPU-readable recording medium.
- the recording medium is a portable medium such as a flexible disk, an optical disk, a magneto-optical disk, and a semiconductor memory.
- the present invention can be applied to a speech recognition apparatus that performs speech recognition processing for recognizing a word string represented by speech from speech.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Probability & Statistics with Applications (AREA)
- Machine Translation (AREA)
Abstract
Description
第1の内容を表す単語列において特定の単語が出現する確率を表す第1の内容別言語モデルと、第2の内容を表す単語列において上記特定の単語が出現する確率を表す第2の内容別言語モデルと、を記憶する内容別言語モデル記憶手段と、
音声に対応する単語列を認識する音声認識処理を行うことにより生成された音声認識仮説に含まれる単語列であって入力された単語列である入力単語列の少なくとも一部である処理対象単語列が表す内容が上記第1の内容である確率を表す第1の確率パラメータと、当該処理対象単語列が表す内容が上記第2の内容である確率を表す第2の確率パラメータと、を取得するとともに、当該取得された第1の確率パラメータと、当該取得された第2の確率パラメータと、上記記憶されている第1の内容別言語モデルと、上記記憶されている第2の内容別言語モデルと、に基づいて、上記音声のうちの上記処理対象単語列に対応する部分に対応する単語列において上記特定の単語が出現する確率を表す言語モデルを作成する言語モデル作成処理を行う言語モデル作成手段と、
を備える。
第1の内容を表す単語列において特定の単語が出現する確率を表す第1の内容別言語モデルと、第2の内容を表す単語列において上記特定の単語が出現する確率を表す第2の内容別言語モデルと、を記憶する内容別言語モデル記憶手段と、
音声に対応する単語列を認識する音声認識処理を行うことにより生成された音声認識仮説に含まれる単語列であって入力された単語列である入力単語列の少なくとも一部である処理対象単語列が表す内容が上記第1の内容である確率を表す第1の確率パラメータと、当該処理対象単語列が表す内容が上記第2の内容である確率を表す第2の確率パラメータと、を取得するとともに、当該取得された第1の確率パラメータと、当該取得された第2の確率パラメータと、上記記憶されている第1の内容別言語モデルと、上記記憶されている第2の内容別言語モデルと、に基づいて、上記音声のうちの上記処理対象単語列に対応する部分に対応する単語列において上記特定の単語が出現する確率を表す言語モデルを作成する言語モデル作成処理を行う言語モデル作成手段と、
上記言語モデル作成手段により作成された言語モデルに基づいて、入力された音声に対応する単語列を認識する音声認識処理を行う音声認識手段と、
を備える。
第1の内容を表す単語列において特定の単語が出現する確率を表す第1の内容別言語モデルと、第2の内容を表す単語列において上記特定の単語が出現する確率を表す第2の内容別言語モデルと、が記憶装置に記憶されている場合に、
音声に対応する単語列を認識する音声認識処理を行うことにより生成された音声認識仮説に含まれる単語列であって入力された単語列である入力単語列の少なくとも一部である処理対象単語列が表す内容が上記第1の内容である確率を表す第1の確率パラメータと、当該処理対象単語列が表す内容が上記第2の内容である確率を表す第2の確率パラメータと、を取得するとともに、当該取得された第1の確率パラメータと、当該取得された第2の確率パラメータと、上記記憶されている第1の内容別言語モデルと、上記記憶されている第2の内容別言語モデルと、に基づいて、上記音声のうちの上記処理対象単語列に対応する部分に対応する単語列において上記特定の単語が出現する確率を表す言語モデルを作成する、方法である。
情報処理装置に、
第1の内容を表す単語列において特定の単語が出現する確率を表す第1の内容別言語モデルと、第2の内容を表す単語列において上記特定の単語が出現する確率を表す第2の内容別言語モデルと、を記憶装置に記憶させる内容別言語モデル記憶処理手段と、
音声に対応する単語列を認識する音声認識処理を行うことにより生成された音声認識仮説に含まれる単語列であって入力された単語列である入力単語列の少なくとも一部である処理対象単語列が表す内容が上記第1の内容である確率を表す第1の確率パラメータと、当該処理対象単語列が表す内容が上記第2の内容である確率を表す第2の確率パラメータと、を取得するとともに、当該取得された第1の確率パラメータと、当該取得された第2の確率パラメータと、上記記憶されている第1の内容別言語モデルと、上記記憶されている第2の内容別言語モデルと、に基づいて、上記音声のうちの上記処理対象単語列に対応する部分に対応する単語列において上記特定の単語が出現する確率を表す言語モデルを作成する言語モデル作成手段と、
を実現させるためのプログラムである。
(構成)
図1を参照しながら、本発明の第1実施形態に係る言語モデル作成装置101について説明する。言語モデル作成装置101は、情報処理装置である。言語モデル作成装置101は、図示しない中央処理装置(CPU;Central Processing Unit)、記憶装置(メモリ及びハードディスク駆動装置(HDD;Hard Disk Drive))、入力装置及び出力装置を備える。
次に、上記のように構成された言語モデル作成装置101の機能について説明する。
この言語モデル作成装置101の機能は、図1に示したように、音声認識仮説入力部(言語モデル作成手段の一部)11と、内容推定部(言語モデル作成手段の一部)12と、言語モデル作成部(言語モデル作成手段の一部)13と、内容モデル記憶部(言語モデル作成手段の一部)14と、内容別言語モデル記憶部(内容別言語モデル記憶手段、内容別言語モデル記憶処理手段、内容別言語モデル記憶処理工程)15と、を含む。この機能は、言語モデル作成装置101のCPUが、記憶装置に記憶されているプログラムを実行することにより実現される。なお、この機能は、論理回路等のハードウェアにより実現されていてもよい。
“Y”は、内容である。即ち、式(1)の右辺P(Y|X)は、処理対象単語列Xが表す内容が内容Yである確率を表す。
また、“Φ(X,Y)”は、処理対象単語列Xの特徴(素性)を表す情報であり、
“Λ”は、素性Φ(X,Y)のそれぞれに対応するCRFにおけるモデルパラメタ(重み値)である。また、“Z”は正規化項である。なお、“exp()”は、eを底とする数値のべき乗を求める関数を示している。
図4の内容候補グラフの円弧Aを例として、該円弧におけるスコアを求める方法の一例を説明する。
次に、本発明の第1実施形態に係る言語モデル作成装置101の作動について、図2に示したフローチャートを参照しながら説明する。
次に、図9を参照しながら、本発明の第2実施形態に係る音声認識装置について説明する。図9は、本発明の第2実施形態にかかる音声認識装置201の機能を示したブロック図である。
言語モデル更新部24は、音声認識仮説入力部11と同様の音声認識仮説入力部241と、内容推定部12と同様の内容推定部242と、言語モデル作成部13と同様の言語モデル作成部243と、内容モデル記憶部14と同様の内容モデル記憶部244と、内容別言語モデル記憶部15と同様の内容別言語モデル記憶部(内容別言語モデル記憶手段、内容別言語モデル記憶処理手段、内容別言語モデル記憶処理工程)245と、を含む。
次に、本発明の第2実施形態に係る音声認識装置の作動について、図10に示したフローチャートを参照しながら説明する。
終了条件が成立している場合、音声認識装置201は、「Yes」と判定して、その時点にて取得された最新の音声認識仮説を音声認識結果として出力する(ステップS27)。
次に、本発明の第3実施形態に係る言語モデル作成装置について図11を参照しながら説明する。
第3実施形態に係る言語モデル作成装置301の機能は、内容別言語モデル記憶部(内容別言語モデル記憶手段)35と、言語モデル作成部(言語モデル作成手段)33と、を含む。
上記取得された第1の確率パラメータが大きくなるほど大きくなる第1の係数を上記第1の内容別言語モデルが表す確率に乗じた値と、上記取得された第2の確率パラメータが大きくなるほど大きくなる第2の係数を上記第2の内容別言語モデルが表す確率に乗じた値と、の和が大きくなるほど、上記音声のうちの上記処理対象単語列に対応する部分に対応する単語列において上記特定の単語が出現する確率が大きくなる上記言語モデルを作成するように構成されることが好適である。
上記言語モデル作成手段は、上記音声認識処理によって認識された単語列の単語間の境界と異なる位置を境界として上記入力単語列を分割した複数の上記処理対象単語列のそれぞれに対して、上記言語モデル作成処理を行うように構成されることが好適である。
上記取得された第1の確率パラメータ及び上記取得された第2の確率パラメータの両方が、予め設定された下限閾値よりも小さい場合、上記取得された第1の確率パラメータ及び上記取得された第2の確率パラメータの両方を同一の値に設定するように構成されることが好適である。
ある処理対象単語列に対して上記取得された第1の確率パラメータが、予め設定された上限閾値よりも大きい場合、その処理対象単語列に隣接する処理対象単語列に対して上記取得された第1の確率パラメータを増加させるように補正するように構成されることが好適である。
上記入力単語列における上記処理対象単語列の位置を表す情報、上記入力単語列が1つの単語を複数含むことを表す情報、上記入力単語列における内容の連接状態を表す情報、及び、上記入力単語列が複数存在する場合における各入力単語列間の関係を表す情報、の少なくとも1つに基づいて上記第1の確率パラメータ及び上記第2の確率パラメータを取得するように構成されることが好適である。
第1の内容を表す単語列において特定の単語が出現する確率を表す第1の内容別言語モデルと、第2の内容を表す単語列において上記特定の単語が出現する確率を表す第2の内容別言語モデルと、を記憶する内容別言語モデル記憶手段と、
音声に対応する単語列を認識する音声認識処理を行うことにより生成された音声認識仮説に含まれる単語列であって入力された単語列である入力単語列の少なくとも一部である処理対象単語列が表す内容が上記第1の内容である確率を表す第1の確率パラメータと、当該処理対象単語列が表す内容が上記第2の内容である確率を表す第2の確率パラメータと、を取得するとともに、当該取得された第1の確率パラメータと、当該取得された第2の確率パラメータと、上記記憶されている第1の内容別言語モデルと、上記記憶されている第2の内容別言語モデルと、に基づいて、上記音声のうちの上記処理対象単語列に対応する部分に対応する単語列において上記特定の単語が出現する確率を表す言語モデルを作成する言語モデル作成処理を行う言語モデル作成手段と、
上記言語モデル作成手段により作成された言語モデルに基づいて、入力された音声に対応する単語列を認識する音声認識処理を行う音声認識手段と、
を備える。
上記取得された第1の確率パラメータが大きくなるほど大きくなる第1の係数を上記第1の内容別言語モデルが表す確率に乗じた値と、上記取得された第2の確率パラメータが大きくなるほど大きくなる第2の係数を上記第2の内容別言語モデルが表す確率に乗じた値と、の和が大きくなるほど、上記音声のうちの上記処理対象単語列に対応する部分に対応する単語列において上記特定の単語が出現する確率が大きくなる上記言語モデルを作成するように構成されることが好適である。
上記音声認識手段は、上記入力された音声に対応する単語列を認識する上記音声認識処理を行うことにより上記入力単語列を生成するように構成され、
上記言語モデル作成手段は、上記音声認識手段により生成された上記入力単語列に基づいて上記言語モデルを作成するように構成され、
上記音声認識手段は、上記言語モデル作成手段により作成された上記言語モデルに基づいて、上記入力された音声に対応する単語列を認識する上記音声認識処理を再度行うように構成されることが好適である。
上記言語モデル作成手段によって作成された言語モデルに基づいて、上記音声認識手段が上記入力された音声に対応する単語列を認識する上記音声認識処理と、上記音声認識手段によって認識された単語列に基づいて、上記言語モデル作成手段が上記言語モデルを作成する上記言語モデル作成処理と、を交互に繰り返す反復処理を実行するように構成されることが好適である。
上記終了条件は、上記音声認識処理を実行した回数が予め設定された閾値回数よりも大きいという条件であることが好適である。
第1の内容を表す単語列において特定の単語が出現する確率を表す第1の内容別言語モデルと、第2の内容を表す単語列において上記特定の単語が出現する確率を表す第2の内容別言語モデルと、が記憶装置に記憶されている場合に、
音声に対応する単語列を認識する音声認識処理を行うことにより生成された音声認識仮説に含まれる単語列であって入力された単語列である入力単語列の少なくとも一部である処理対象単語列が表す内容が上記第1の内容である確率を表す第1の確率パラメータと、当該処理対象単語列が表す内容が上記第2の内容である確率を表す第2の確率パラメータと、を取得するとともに、当該取得された第1の確率パラメータと、当該取得された第2の確率パラメータと、上記記憶されている第1の内容別言語モデルと、上記記憶されている第2の内容別言語モデルと、に基づいて、上記音声のうちの上記処理対象単語列に対応する部分に対応する単語列において上記特定の単語が出現する確率を表す言語モデルを作成する、方法である。
上記取得された第1の確率パラメータが大きくなるほど大きくなる第1の係数を上記第1の内容別言語モデルが表す確率に乗じた値と、上記取得された第2の確率パラメータが大きくなるほど大きくなる第2の係数を上記第2の内容別言語モデルが表す確率に乗じた値と、の和が大きくなるほど、上記音声のうちの上記処理対象単語列に対応する部分に対応する単語列において上記特定の単語が出現する確率が大きくなる上記言語モデルを作成するように構成されることが好適である。
情報処理装置に、
第1の内容を表す単語列において特定の単語が出現する確率を表す第1の内容別言語モデルと、第2の内容を表す単語列において上記特定の単語が出現する確率を表す第2の内容別言語モデルと、を記憶装置に記憶させる内容別言語モデル記憶処理手段と、
音声に対応する単語列を認識する音声認識処理を行うことにより生成された音声認識仮説に含まれる単語列であって入力された単語列である入力単語列の少なくとも一部である処理対象単語列が表す内容が上記第1の内容である確率を表す第1の確率パラメータと、当該処理対象単語列が表す内容が上記第2の内容である確率を表す第2の確率パラメータと、を取得するとともに、当該取得された第1の確率パラメータと、当該取得された第2の確率パラメータと、上記記憶されている第1の内容別言語モデルと、上記記憶されている第2の内容別言語モデルと、に基づいて、上記音声のうちの上記処理対象単語列に対応する部分に対応する単語列において上記特定の単語が出現する確率を表す言語モデルを作成する言語モデル作成手段と、
を実現させるためのプログラムである。
上記取得された第1の確率パラメータが大きくなるほど大きくなる第1の係数を上記第1の内容別言語モデルが表す確率に乗じた値と、上記取得された第2の確率パラメータが大きくなるほど大きくなる第2の係数を上記第2の内容別言語モデルが表す確率に乗じた値と、の和が大きくなるほど、上記音声のうちの上記処理対象単語列に対応する部分に対応する単語列において上記特定の単語が出現する確率が大きくなる上記言語モデルを作成するように構成されることが好適である。
12 内容推定部
13 言語モデル作成部
14 内容モデル記憶部
15 内容別言語モデル記憶部
21 音声認識部
22 音声認識モデル記憶部
24 言語モデル更新部
33 言語モデル作成部
35 内容別言語モデル記憶部
101 言語モデル作成装置
201 音声認識装置
241 音声認識仮説入力部
242 内容推定部
243 言語モデル作成部
244 内容モデル記憶部
245 内容別言語モデル記憶部
301 言語モデル作成装置
Claims (22)
- 第1の内容を表す単語列において特定の単語が出現する確率を表す第1の内容別言語モデルと、第2の内容を表す単語列において前記特定の単語が出現する確率を表す第2の内容別言語モデルと、を記憶する内容別言語モデル記憶手段と、
音声に対応する単語列を認識する音声認識処理を行うことにより生成された音声認識仮説に含まれる単語列であって入力された単語列である入力単語列の少なくとも一部である処理対象単語列が表す内容が前記第1の内容である確率を表す第1の確率パラメータと、当該処理対象単語列が表す内容が前記第2の内容である確率を表す第2の確率パラメータと、を取得するとともに、当該取得された第1の確率パラメータと、当該取得された第2の確率パラメータと、前記記憶されている第1の内容別言語モデルと、前記記憶されている第2の内容別言語モデルと、に基づいて、前記音声のうちの前記処理対象単語列に対応する部分に対応する単語列において前記特定の単語が出現する確率を表す言語モデルを作成する言語モデル作成処理を行う言語モデル作成手段と、
を備える言語モデル作成装置。 - 請求項1に記載の言語モデル作成装置であって、
前記言語モデル作成手段は、
前記取得された第1の確率パラメータが大きくなるほど大きくなる第1の係数を前記第1の内容別言語モデルが表す確率に乗じた値と、前記取得された第2の確率パラメータが大きくなるほど大きくなる第2の係数を前記第2の内容別言語モデルが表す確率に乗じた値と、の和が大きくなるほど、前記音声のうちの前記処理対象単語列に対応する部分に対応する単語列において前記特定の単語が出現する確率が大きくなる前記言語モデルを作成するように構成された言語モデル作成装置。 - 請求項1又は請求項2に記載の言語モデル作成装置であって、
前記言語モデル作成手段は、前記音声認識処理によって認識された単語列の単語間の境界を境界として前記入力単語列を分割した複数の前記処理対象単語列のそれぞれに対して、前記言語モデル作成処理を行うように構成された言語モデル作成装置。 - 請求項1又は請求項2に記載の言語モデル作成装置であって、
前記言語モデル作成手段は、前記音声認識処理によって認識された単語列の単語間の境界と異なる位置を境界として前記入力単語列を分割した複数の前記処理対象単語列のそれぞれに対して、前記言語モデル作成処理を行うように構成された言語モデル作成装置。 - 請求項3又は請求項4に記載の言語モデル作成装置であって、
前記言語モデル作成手段は、
前記取得された第1の確率パラメータ及び前記取得された第2の確率パラメータの両方が、予め設定された下限閾値よりも小さい場合、前記取得された第1の確率パラメータ及び前記取得された第2の確率パラメータの両方を同一の値に設定するように構成された言語モデル作成装置。 - 請求項3乃至請求項5のいずれか一項に記載の言語モデル作成装置であって、
前記言語モデル作成手段は、
ある処理対象単語列に対して前記取得された第1の確率パラメータが、予め設定された上限閾値よりも大きい場合、その処理対象単語列に隣接する処理対象単語列に対して前記取得された第1の確率パラメータを増加させるように補正するように構成された言語モデル作成装置。 - 請求項1乃至請求項6のいずれか一項に記載の言語モデル作成装置であって、
前記言語モデル作成手段は、条件付確率場の理論に基づいて前記第1の確率パラメータ及び前記第2の確率パラメータを取得するように構成された言語モデル作成装置。 - 請求項1乃至請求項7のいずれか一項に記載の言語モデル作成装置であって、
前記言語モデル作成手段は、前記処理対象単語列に含まれる単語の属性を表す単語属性情報、及び、前記処理対象単語列を音声から認識する音声認識処理を行う際に取得された音声認識処理情報、の少なくとも1つに基づいて前記第1の確率パラメータ及び前記第2の確率パラメータを取得するように構成された言語モデル作成装置。 - 請求項8に記載の言語モデル作成装置であって、
前記単語属性情報は、単語表層を表す情報、読みを表す情報、及び、品詞を表す情報の少なくとも1つを含む言語モデル作成装置。 - 請求項8又は請求項9に記載の言語モデル作成装置であって、
前記音声認識処理情報は、前記音声認識処理による認識結果の信頼度である認識信頼度を表す情報、1つの音が継続する時間である継続時間長を表す情報、及び、先行する無音の有無を表す情報の少なくとも1つを含む言語モデル作成装置。 - 請求項1乃至請求項10のいずれか一項に記載の言語モデル作成装置であって、
前記言語モデル作成手段は、
前記入力単語列における前記処理対象単語列の位置を表す情報、前記入力単語列が1つの単語を複数含むことを表す情報、前記入力単語列における内容の連接状態を表す情報、及び、前記入力単語列が複数存在する場合における各入力単語列間の関係を表す情報、の少なくとも1つに基づいて前記第1の確率パラメータ及び前記第2の確率パラメータを取得するように構成された言語モデル作成装置。 - 第1の内容を表す単語列において特定の単語が出現する確率を表す第1の内容別言語モデルと、第2の内容を表す単語列において前記特定の単語が出現する確率を表す第2の内容別言語モデルと、を記憶する内容別言語モデル記憶手段と、
音声に対応する単語列を認識する音声認識処理を行うことにより生成された音声認識仮説に含まれる単語列であって入力された単語列である入力単語列の少なくとも一部である処理対象単語列が表す内容が前記第1の内容である確率を表す第1の確率パラメータと、当該処理対象単語列が表す内容が前記第2の内容である確率を表す第2の確率パラメータと、を取得するとともに、当該取得された第1の確率パラメータと、当該取得された第2の確率パラメータと、前記記憶されている第1の内容別言語モデルと、前記記憶されている第2の内容別言語モデルと、に基づいて、前記音声のうちの前記処理対象単語列に対応する部分に対応する単語列において前記特定の単語が出現する確率を表す言語モデルを作成する言語モデル作成処理を行う言語モデル作成手段と、
前記言語モデル作成手段により作成された言語モデルに基づいて、入力された音声に対応する単語列を認識する音声認識処理を行う音声認識手段と、
を備える音声認識装置。 - 請求項12に記載の音声認識装置であって、
前記言語モデル作成手段は、
前記取得された第1の確率パラメータが大きくなるほど大きくなる第1の係数を前記第1の内容別言語モデルが表す確率に乗じた値と、前記取得された第2の確率パラメータが大きくなるほど大きくなる第2の係数を前記第2の内容別言語モデルが表す確率に乗じた値と、の和が大きくなるほど、前記音声のうちの前記処理対象単語列に対応する部分に対応する単語列において前記特定の単語が出現する確率が大きくなる前記言語モデルを作成するように構成された音声認識装置。 - 請求項12又は請求項13に記載の音声認識装置であって、
前記音声認識手段は、前記入力された音声に対応する単語列を認識する前記音声認識処理を行うことにより前記入力単語列を生成するように構成され、
前記言語モデル作成手段は、前記音声認識手段により生成された前記入力単語列に基づいて前記言語モデルを作成するように構成され、
前記音声認識手段は、前記言語モデル作成手段により作成された前記言語モデルに基づいて、前記入力された音声に対応する単語列を認識する前記音声認識処理を再度行うように構成された音声認識装置。 - 請求項12乃至請求項14のいずれか一項に記載の音声認識装置であって、
前記言語モデル作成手段によって作成された言語モデルに基づいて、前記音声認識手段が前記入力された音声に対応する単語列を認識する前記音声認識処理と、前記音声認識手段によって認識された単語列に基づいて、前記言語モデル作成手段が前記言語モデルを作成する前記言語モデル作成処理と、を交互に繰り返す反復処理を実行するように構成された音声認識装置。 - 請求項15に記載の音声認識装置であって、
所定の終了条件が成立した場合、前記反復処理を終了するように構成された音声認識装置。 - 請求項16に記載の音声認識装置であって、
前記終了条件は、前回の前記音声認識処理により認識された単語列と、今回の前記音声認識処理により認識された単語列と、が一致しているという条件である音声認識装置。 - 請求項16に記載の音声認識装置であって、
前記終了条件は、前記音声認識処理を実行した回数が予め設定された閾値回数よりも大きいという条件である音声認識装置。 - 第1の内容を表す単語列において特定の単語が出現する確率を表す第1の内容別言語モデルと、第2の内容を表す単語列において前記特定の単語が出現する確率を表す第2の内容別言語モデルと、が記憶装置に記憶されている場合に、
音声に対応する単語列を認識する音声認識処理を行うことにより生成された音声認識仮説に含まれる単語列であって入力された単語列である入力単語列の少なくとも一部である処理対象単語列が表す内容が前記第1の内容である確率を表す第1の確率パラメータと、当該処理対象単語列が表す内容が前記第2の内容である確率を表す第2の確率パラメータと、を取得するとともに、当該取得された第1の確率パラメータと、当該取得された第2の確率パラメータと、前記記憶されている第1の内容別言語モデルと、前記記憶されている第2の内容別言語モデルと、に基づいて、前記音声のうちの前記処理対象単語列に対応する部分に対応する単語列において前記特定の単語が出現する確率を表す言語モデルを作成する、言語モデル作成方法。 - 請求項19に記載の言語モデル作成方法であって、
前記取得された第1の確率パラメータが大きくなるほど大きくなる第1の係数を前記第1の内容別言語モデルが表す確率に乗じた値と、前記取得された第2の確率パラメータが大きくなるほど大きくなる第2の係数を前記第2の内容別言語モデルが表す確率に乗じた値と、の和が大きくなるほど、前記音声のうちの前記処理対象単語列に対応する部分に対応する単語列において前記特定の単語が出現する確率が大きくなる前記言語モデルを作成するように構成された言語モデル作成方法。 - 情報処理装置に、
第1の内容を表す単語列において特定の単語が出現する確率を表す第1の内容別言語モデルと、第2の内容を表す単語列において前記特定の単語が出現する確率を表す第2の内容別言語モデルと、を記憶装置に記憶させる内容別言語モデル記憶処理手段と、
音声に対応する単語列を認識する音声認識処理を行うことにより生成された音声認識仮説に含まれる単語列であって入力された単語列である入力単語列の少なくとも一部である処理対象単語列が表す内容が前記第1の内容である確率を表す第1の確率パラメータと、当該処理対象単語列が表す内容が前記第2の内容である確率を表す第2の確率パラメータと、を取得するとともに、当該取得された第1の確率パラメータと、当該取得された第2の確率パラメータと、前記記憶されている第1の内容別言語モデルと、前記記憶されている第2の内容別言語モデルと、に基づいて、前記音声のうちの前記処理対象単語列に対応する部分に対応する単語列において前記特定の単語が出現する確率を表す言語モデルを作成する言語モデル作成手段と、
を実現させるための言語モデル作成プログラム。 - 請求項21に記載の言語モデル作成プログラムであって、
前記言語モデル作成手段は、
前記取得された第1の確率パラメータが大きくなるほど大きくなる第1の係数を前記第1の内容別言語モデルが表す確率に乗じた値と、前記取得された第2の確率パラメータが大きくなるほど大きくなる第2の係数を前記第2の内容別言語モデルが表す確率に乗じた値と、の和が大きくなるほど、前記音声のうちの前記処理対象単語列に対応する部分に対応する単語列において前記特定の単語が出現する確率が大きくなる前記言語モデルを作成するように構成された言語モデル作成プログラム。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US13/129,721 US9043209B2 (en) | 2008-11-28 | 2009-09-03 | Language model creation device |
JP2010540302A JP5598331B2 (ja) | 2008-11-28 | 2009-09-03 | 言語モデル作成装置 |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2008304564 | 2008-11-28 | ||
JP2008-304564 | 2008-11-28 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2010061507A1 true WO2010061507A1 (ja) | 2010-06-03 |
Family
ID=42225399
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2009/004341 WO2010061507A1 (ja) | 2008-11-28 | 2009-09-03 | 言語モデル作成装置 |
Country Status (3)
Country | Link |
---|---|
US (1) | US9043209B2 (ja) |
JP (1) | JP5598331B2 (ja) |
WO (1) | WO2010061507A1 (ja) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103918026A (zh) * | 2011-11-01 | 2014-07-09 | 谷歌公司 | 用于增量地生成的语音识别假设的增强的稳定性预测 |
CN109712607A (zh) * | 2018-12-30 | 2019-05-03 | 联想(北京)有限公司 | 一种处理方法、装置及电子设备 |
Families Citing this family (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090313017A1 (en) * | 2006-07-07 | 2009-12-17 | Satoshi Nakazawa | Language model update device, language Model update method, and language model update program |
JP2010008601A (ja) * | 2008-06-25 | 2010-01-14 | Fujitsu Ltd | 案内情報表示装置、案内情報表示方法及びプログラム |
EP2339576B1 (en) | 2009-12-23 | 2019-08-07 | Google LLC | Multi-modal input on an electronic device |
US11416214B2 (en) | 2009-12-23 | 2022-08-16 | Google Llc | Multi-modal input on an electronic device |
US8352245B1 (en) | 2010-12-30 | 2013-01-08 | Google Inc. | Adjusting language models |
US8296142B2 (en) | 2011-01-21 | 2012-10-23 | Google Inc. | Speech recognition using dock context |
US9672818B2 (en) | 2013-04-18 | 2017-06-06 | Nuance Communications, Inc. | Updating population language models based on changes made by user clusters |
US9842592B2 (en) | 2014-02-12 | 2017-12-12 | Google Inc. | Language models using non-linguistic context |
US9412365B2 (en) | 2014-03-24 | 2016-08-09 | Google Inc. | Enhanced maximum entropy models |
US20150309984A1 (en) * | 2014-04-25 | 2015-10-29 | Nuance Communications, Inc. | Learning language models from scratch based on crowd-sourced user text input |
KR102389313B1 (ko) | 2015-01-16 | 2022-04-21 | 삼성전자주식회사 | 문법 모델을 이용하여 음성인식을 수행하는 방법 및 디바이스 |
US10134394B2 (en) | 2015-03-20 | 2018-11-20 | Google Llc | Speech recognition using log-linear model |
KR20170046291A (ko) * | 2015-10-21 | 2017-05-02 | 삼성전자주식회사 | 전자 기기, 그의 음향 모델 적응 방법 및 음성 인식 시스템 |
US9978367B2 (en) | 2016-03-16 | 2018-05-22 | Google Llc | Determining dialog states for language models |
US10832664B2 (en) | 2016-08-19 | 2020-11-10 | Google Llc | Automated speech recognition using language models that selectively use domain-specific model components |
US10311860B2 (en) | 2017-02-14 | 2019-06-04 | Google Llc | Language model biasing system |
CN109325227A (zh) * | 2018-09-14 | 2019-02-12 | 北京字节跳动网络技术有限公司 | 用于生成修正语句的方法和装置 |
WO2020112789A1 (en) | 2018-11-30 | 2020-06-04 | Google Llc | Speech processing |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH11259084A (ja) * | 1998-03-16 | 1999-09-24 | Kdd Corp | クラスタ混合分布言語モデル作成装置 |
JP2004198597A (ja) * | 2002-12-17 | 2004-07-15 | Advanced Telecommunication Research Institute International | 音声認識装置および文分類装置としてコンピュータを動作させるコンピュータプログラム、階層化された言語モデルを作成する方法を実現する様にコンピュータを動作させるコンピュータプログラム、および記憶媒体 |
JP2005284209A (ja) * | 2004-03-31 | 2005-10-13 | Kddi Corp | 音声認識方式 |
WO2005122143A1 (ja) * | 2004-06-08 | 2005-12-22 | Matsushita Electric Industrial Co., Ltd. | 音声認識装置および音声認識方法 |
WO2008001485A1 (fr) * | 2006-06-26 | 2008-01-03 | Nec Corporation | système de génération de modèles de langue, procédé de génération de modèles de langue et programme de génération de modèles de langue |
WO2008004666A1 (fr) * | 2006-07-07 | 2008-01-10 | Nec Corporation | Dispositif, procédé et programme de reconnaissance vocale |
Family Cites Families (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH05197389A (ja) * | 1991-08-13 | 1993-08-06 | Toshiba Corp | 音声認識装置 |
US5541836A (en) * | 1991-12-30 | 1996-07-30 | At&T Corp. | Word disambiguation apparatus and methods |
US5873056A (en) * | 1993-10-12 | 1999-02-16 | The Syracuse University | Natural language processing system for semantic vector representation which accounts for lexical ambiguity |
US5842163A (en) * | 1995-06-21 | 1998-11-24 | Sri International | Method and apparatus for computing likelihood and hypothesizing keyword appearance in speech |
AU6849196A (en) * | 1995-08-16 | 1997-03-19 | Syracuse University | Multilingual document retrieval system and method using semantic vector matching |
DE19533541C1 (de) * | 1995-09-11 | 1997-03-27 | Daimler Benz Aerospace Ag | Verfahren zur automatischen Steuerung eines oder mehrerer Geräte durch Sprachkommandos oder per Sprachdialog im Echtzeitbetrieb und Vorrichtung zum Ausführen des Verfahrens |
US5828999A (en) * | 1996-05-06 | 1998-10-27 | Apple Computer, Inc. | Method and system for deriving a large-span semantic language model for large-vocabulary recognition systems |
US7319957B2 (en) * | 2004-02-11 | 2008-01-15 | Tegic Communications, Inc. | Handwriting and voice input with automatic correction |
WO2001046945A1 (en) * | 1999-12-20 | 2001-06-28 | British Telecommunications Public Limited Company | Learning of dialogue states and language model of spoken information system |
JP3426176B2 (ja) * | 1999-12-27 | 2003-07-14 | インターナショナル・ビジネス・マシーンズ・コーポレーション | 音声認識装置、方法、コンピュータ・システム及び記憶媒体 |
US7072827B1 (en) * | 2000-06-29 | 2006-07-04 | International Business Machines Corporation | Morphological disambiguation |
JP4105841B2 (ja) * | 2000-07-11 | 2008-06-25 | インターナショナル・ビジネス・マシーンズ・コーポレーション | 音声認識方法、音声認識装置、コンピュータ・システムおよび記憶媒体 |
DE10038517A1 (de) * | 2000-08-08 | 2002-02-21 | Philips Corp Intellectual Pty | Automatissche Erkennung von Unternehmensnamen in sprachlichen Äußerungen |
JP2002229589A (ja) | 2001-01-29 | 2002-08-16 | Mitsubishi Electric Corp | 音声認識装置 |
US7133827B1 (en) * | 2002-02-06 | 2006-11-07 | Voice Signal Technologies, Inc. | Training speech recognition word models from word samples synthesized by Monte Carlo techniques |
US7236931B2 (en) * | 2002-05-01 | 2007-06-26 | Usb Ag, Stamford Branch | Systems and methods for automatic acoustic speaker adaptation in computer-assisted transcription systems |
US20040111259A1 (en) * | 2002-12-10 | 2004-06-10 | Miller Edward S. | Speech recognition system having an application program interface |
EP1450350A1 (en) * | 2003-02-20 | 2004-08-25 | Sony International (Europe) GmbH | Method for Recognizing Speech with attributes |
US7487094B1 (en) * | 2003-06-20 | 2009-02-03 | Utopy, Inc. | System and method of call classification with context modeling based on composite words |
WO2005020093A1 (en) * | 2003-08-21 | 2005-03-03 | Idilia Inc. | Internet searching using semantic disambiguation and expansion |
US7529657B2 (en) * | 2004-09-24 | 2009-05-05 | Microsoft Corporation | Configurable parameters for grammar authoring for speech recognition and natural language understanding |
US7548849B2 (en) * | 2005-04-29 | 2009-06-16 | Research In Motion Limited | Method for generating text that meets specified characteristics in a handheld electronic device and a handheld electronic device incorporating the same |
US7681147B2 (en) * | 2005-12-13 | 2010-03-16 | Yahoo! Inc. | System for determining probable meanings of inputted words |
US8009819B2 (en) * | 2007-03-26 | 2011-08-30 | Nuance Communications, Inc. | Semi-supervised training of destination map for call handling applications |
-
2009
- 2009-09-03 JP JP2010540302A patent/JP5598331B2/ja active Active
- 2009-09-03 WO PCT/JP2009/004341 patent/WO2010061507A1/ja active Application Filing
- 2009-09-03 US US13/129,721 patent/US9043209B2/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH11259084A (ja) * | 1998-03-16 | 1999-09-24 | Kdd Corp | クラスタ混合分布言語モデル作成装置 |
JP2004198597A (ja) * | 2002-12-17 | 2004-07-15 | Advanced Telecommunication Research Institute International | 音声認識装置および文分類装置としてコンピュータを動作させるコンピュータプログラム、階層化された言語モデルを作成する方法を実現する様にコンピュータを動作させるコンピュータプログラム、および記憶媒体 |
JP2005284209A (ja) * | 2004-03-31 | 2005-10-13 | Kddi Corp | 音声認識方式 |
WO2005122143A1 (ja) * | 2004-06-08 | 2005-12-22 | Matsushita Electric Industrial Co., Ltd. | 音声認識装置および音声認識方法 |
WO2008001485A1 (fr) * | 2006-06-26 | 2008-01-03 | Nec Corporation | système de génération de modèles de langue, procédé de génération de modèles de langue et programme de génération de modèles de langue |
WO2008004666A1 (fr) * | 2006-07-07 | 2008-01-10 | Nec Corporation | Dispositif, procédé et programme de reconnaissance vocale |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103918026A (zh) * | 2011-11-01 | 2014-07-09 | 谷歌公司 | 用于增量地生成的语音识别假设的增强的稳定性预测 |
CN109712607A (zh) * | 2018-12-30 | 2019-05-03 | 联想(北京)有限公司 | 一种处理方法、装置及电子设备 |
Also Published As
Publication number | Publication date |
---|---|
US20110231183A1 (en) | 2011-09-22 |
JP5598331B2 (ja) | 2014-10-01 |
US9043209B2 (en) | 2015-05-26 |
JPWO2010061507A1 (ja) | 2012-04-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP5598331B2 (ja) | 言語モデル作成装置 | |
US11900915B2 (en) | Multi-dialect and multilingual speech recognition | |
US10210862B1 (en) | Lattice decoding and result confirmation using recurrent neural networks | |
US10037758B2 (en) | Device and method for understanding user intent | |
JP5440177B2 (ja) | 単語カテゴリ推定装置、単語カテゴリ推定方法、音声認識装置、音声認識方法、プログラム、および記録媒体 | |
US8155956B2 (en) | Voice query extension method and system | |
JP3782943B2 (ja) | 音声認識装置、コンピュータ・システム、音声認識方法、プログラムおよび記録媒体 | |
US8311825B2 (en) | Automatic speech recognition method and apparatus | |
US8849668B2 (en) | Speech recognition apparatus and method | |
JP5861649B2 (ja) | モデル適応化装置、モデル適応化方法およびモデル適応化用プログラム | |
US20070100814A1 (en) | Apparatus and method for detecting named entity | |
JP2005010691A (ja) | 音声認識装置、音声認識方法、会話制御装置、会話制御方法及びこれらのためのプログラム | |
US6801891B2 (en) | Speech processing system | |
JP4930379B2 (ja) | 類似文検索方法、類似文検索システム及び類似文検索用プログラム | |
JP2014521115A (ja) | 信頼度計算の方法及び装置 | |
JP6047364B2 (ja) | 音声認識装置、誤り修正モデル学習方法、及びプログラム | |
JP2001249684A (ja) | 音声認識装置および音声認識方法、並びに記録媒体 | |
US10360904B2 (en) | Methods and apparatus for speech recognition using a garbage model | |
US11227580B2 (en) | Speech recognition accuracy deterioration factor estimation device, speech recognition accuracy deterioration factor estimation method, and program | |
JP6031316B2 (ja) | 音声認識装置、誤り修正モデル学習方法、及びプログラム | |
JP2008241970A (ja) | 話者適応装置、話者適応方法及び話者適応プログラム | |
JP2010224153A (ja) | 音声対話装置及びプログラム | |
Rybach et al. | On lattice generation for large vocabulary speech recognition | |
JP6276516B2 (ja) | 辞書作成装置、及び辞書作成プログラム | |
JP5585111B2 (ja) | 発話内容推定装置、言語モデル作成装置、それに用いる方法およびプログラム |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 09828757 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 13129721 Country of ref document: US |
|
ENP | Entry into the national phase |
Ref document number: 2010540302 Country of ref document: JP Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 09828757 Country of ref document: EP Kind code of ref document: A1 |