JP4089861B2 - Voice recognition text input device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To solve the problem that it was difficult before to recognize a technical sentence including special terms and phrasings. SOLUTION: This device is equipped with a language model 3, a language model 3b which has a vocabulary characteristic of a user, a speech input means 2 which inputs speech, a speech character converting means 4 which converts the speech into characters by computing the likelihood of a word sequence by referring to the language model and user language model, a text buffer 5 which temporarily stores the word sequence of the output characters of the speech character converting means, a display means 6 which displays the characters temporarily stored in the text buffer, a correcting means 8 which generates a user text by correcting the characters displayed by the display means, and a user language model learning means 10 which analyzes the user text and adds the vocabulary characteristic of the user to the user language model to update it. Consequently, the vocabulary and expression characteristic of the user can be added to the user language model and recognition performance can be improved corresponding a speaking variation characteristic of the user.

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a speech recognition text input device having a business automatic adaptation function for recognizing speech and inputting text.
[0002]
[Prior art]
Expectations for document creation by voice input are extremely high because convenience and special training are not required, and Japanese text input software by voice recognition is commercially available from various companies and attracts attention.
[0003]
A conventional speech recognition text input device will be described with reference to the drawings. FIG. 23 is a diagram showing a general block configuration of a conventional speech recognition text input device realized by operating the above software on a computer.
[0004]
In FIG. 23, 1 is a voice, 2 is a voice input means for capturing the input voice 1, 3 is a language model composed of N-grams of words, for example, 4 is an input voice acquired by the voice input means 2 with reference to the language model 3 1 is a speech character converting means for converting 1 into a word string, 5 is a text buffer for temporarily storing an output word string of the speech character converting means 4, 6 is a display means for displaying characters stored in the text buffer 5, and 8 is a user. Correction means 9 for correcting characters in the text buffer 5 by the operation 7, and user text created by the user extracted from the text buffer 5. As the language model 3, Ngram (for example, N = 3) based on word chain statistics is used.
[0005]
Next, the operation of the conventional speech recognition text input device will be described with reference to the drawings.
[0006]
In the above configuration, the user's voice 1 is converted into a word string according to the word chain information stored in the language model 3 by the phonetic character conversion means 4 and temporarily stored in the text buffer 5, and at the same time by the display means 6. Displayed to the user.
[0007]
The user can correct the recognition error in the text temporarily stored in the text buffer 5 based on the display on the display means 6 and finally obtain the desired user text 9.
[0008]
However, in such a conventional speech recognition text input device, since the language model 3 is created from a large number of general text examples such as newspapers and WEB, it is difficult to recognize specialized text including special terms and phrases. Moreover, the recognition performance is insufficient simply by registering technical terms. This problem is also described in Japanese Patent Application Laid-Open No. 2000-250584.
[0009]
On the other hand, it is possible to create a language model corresponding to the field of specialized work from sentence examples accumulated in specialized work, but generally there is little sentence accumulation (about 10,000 sentences at most) for each specialized work, and it is effective. Language model creation is considered difficult. For example, when an N-gram of a word is used as a language model, it is said that 1 to 10 million or more example sentences are necessary when N = 3.
[0010]
For this reason, in the speech recognition apparatus disclosed in Japanese Patent Application Laid-Open No. 10-198395, data for specific tasks (tasks) is prepared in advance from a plurality of tasks, and these are used as prior knowledge, and tasks for which text is input. A language model adapted to is generated.
[0011]
However, when this method is applied to, for example, a highly specialized task (for example, a specific medical treatment subject), the text data of the specific task for learning prepared in advance is the terminology or task of the task to be applied. There are many cases in which specific expressions (word chains) are not included, and there is a problem that these text data for learning have little value as prior knowledge.
[0012]
On the other hand, a syntax-driven speech recognition method that uses a language model that expresses a sentence to be recognized by a network of words has been known. However, in this case, a small amount is required when dealing with a highly specialized field. There is a problem that it is necessary to create vocabulary and grammar from a sentence example by spending a great deal of labor and time.
[0013]
In addition, in this way, the language model made from a small amount of text of the target business has solved the problem that the recognition performance is significantly deteriorated with respect to fluctuations of expressions that are not included in the sentence examples and fluctuations of expressions for each individual. Not.
[0014]
For these reasons, the practical use of professional dictation software has not progressed.
[0015]
[Problems to be solved by the invention]
In the conventional speech recognition text input device as described above, the language model is created from a large number of general text examples such as newspapers and WEB, so that it is difficult to recognize special text including special terms and phrases. was there.
[0016]
The present invention has been made to solve the above-described problems, and effectively generates a language model that is adapted to a specific target job and has a high tolerance for utterance variations, and is generated while the system is in use. The purpose is to obtain a speech recognition text input device that has high recognition performance and sufficient reliability for professionals with improved freedom of speech for unknown expressions, etc., and excellent adaptability to various tasks.
[0017]
[Means for Solving the Problems]
According to a first aspect of the present invention, there is provided a speech recognition text input device comprising: a language model composed of N-grams of words; and a user-specific vocabulary having N-grams of words. User-specific vocabulary and expressions can be added from user text created by user operations. A speech recognition process with reference to an acoustic model composed of a phoneme environment-dependent phoneme HMM, the language model, and the user language model. Calculate the likelihood of the word string with the respective language model and based on the larger likelihood, A speech character converting means for converting an input speech into a word string having the maximum likelihood, a text buffer for temporarily storing the word string converted by the speech character converting means, and characters temporarily stored in the text buffer are displayed. A display unit, a correction unit that corrects the characters displayed on the display unit as needed by a user operation to create a user text, and an analysis dictionary dedicated to text analysis that is a dictionary specialized for the target task. The user text is analyzed with reference to the analysis dictionary, divided into words constituting the user text, and a set of consecutive N words (N is an integer) is completely extracted from the divided word sequence. Then, the extracted N word sets are counted to count the number of appearances of each N word set, and the number of appearances of the N word set is added to the user language model to add a user-specific vocabulary, User-specific It comprises a user language model learning means for learning expressions.
[0018]
The speech recognition text input device according to claim 2 of the present invention includes a learning start notification means for generating a learning start notification in response to a user operation, and a learning start notification from the learning start notification means. And a control unit that activates the user language model learning unit.
[0019]
According to a third aspect of the present invention, there is provided a speech recognition text input device comprising: a detecting unit that detects that a character has been corrected by the correcting unit; and the user language model learning after the character correction is detected by the detecting unit. And a control means for activating the means.
[0020]
In the speech recognition text input device according to claim 4 of the present invention, the language model is a target business language model learned from a sentence example of the target business.
[0021]
The speech recognition text input device according to claim 5 of the present invention further includes a background language model learned from general sentence examples.
[0022]
The speech recognition text input device according to claim 6 of the present invention adds a sentence example storage means for storing a sentence example for use in learning of the target business language model, and a sentence example based on the user text to the sentence example storage means. It further comprises sentence example adding means and language model learning means for creating the target business language model using the sentence examples stored in the sentence example storage means.
[0023]
According to a seventh aspect of the present invention, there is provided a speech recognition text input device, an analyzed sentence example storage means for storing an analyzed sentence example for use in learning of the target business language model, and an analysis result output by analyzing the user text. Using sentence analysis means, analyzed sentence example addition means for additionally storing the analyzed sentence example analyzed by the sentence analysis means in the analyzed sentence example storage means, and analyzed sentence examples stored in the analyzed sentence example storage means Language model learning means for creating the target business language model, and the user language model learning means creates the user language model using the analyzed sentence example from the sentence analysis means.
[0024]
According to an eighth aspect of the present invention, there is provided a speech recognition text input device for extracting an unknown word that passes a syllable recognition result of a user utterance as a pronunciation together with the unknown word to the text analysis means when an unknown word exists in the user text. Means are further provided.
[0025]
In the speech recognition text input device according to claim 9 of the present invention, the phonetic character conversion unit classifies the words of the language model into classes, and estimates the probability of arranging words that are not included in the sentence examples from the characteristics of the class arrangement. Language probability estimation means, preliminary search means for creating a word lattice using the probabilities estimated by the language probability estimation means, and search for word strings from the word lattice using probabilities estimated by the language probability estimation means And a word string search means.
[0026]
In the speech recognition text input device according to claim 10 of the present invention, the phonetic character conversion unit classifies the words of the language model into classes, and estimates the probability of arranging words that are not included in the sentence examples from the characteristics of the class arrangement. Language probability estimation means, difference model storing recognition error tendency of basic symbol strings, basic symbol string recognition means for recognizing basic symbol strings without referring to the language model, and from the recognized basic symbol strings And a word string search means for searching for a word string using the language model via the difference model and the language probability estimation means.
[0027]
According to an eleventh aspect of the present invention, there is provided a speech recognition text input device that includes speech input means for inputting speech, a text buffer for temporarily storing a word string converted from speech to characters, and displaying characters temporarily stored in the text buffer. And a client computer having a display unit for correcting the character displayed on the display unit as needed by a user operation to create a user text, and connected to the client computer via a network, and an N-gram of a word A language model that consists of N-grams of words with user-specific vocabulary User-specific vocabulary and expressions can be added from user text created by user operations. A user language model, an acoustic model composed of a phoneme environment-dependent phoneme HMM, the language model, and the user language model, Calculate the likelihood of the word string with the respective language model and based on the larger likelihood, A speech character converting means for converting the input speech into a word string having the maximum likelihood; and an analysis dictionary dedicated to text analysis which is a dictionary specialized for the target business, and the user text is referred to the analysis dictionary. Are divided into words constituting the user text, and all consecutive sets of N words (N is an integer) are extracted from the divided word strings, and the extracted sets of N words are aggregated to obtain N Count the number of occurrences for each set of words, add the number of occurrences of a set of N words to the user language model, User-specific And a host computer having user language model learning means for learning expressions.
[0028]
DETAILED DESCRIPTION OF THE INVENTION
Embodiment 1 FIG.
A speech recognition text input device according to Embodiment 1 of the present invention will be described with reference to the drawings. FIG. 1 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 1 of the present invention. In addition, in each figure, the same code | symbol shows the same or equivalent part.
[0029]
In FIG. 1, 2 is a voice input means for capturing the voice 1 spoken by the user, 3 is a language model, 3b is a user language model, 4 is acquired by the voice input means 2 with reference to the language model 3 and the user language model 3b. A voice character converting means for converting the input voice 1 into a word string, 5 a text buffer for temporarily storing the output word string of the voice character converting means 4, and 6 a display means for displaying the characters stored in the text buffer 5. 8 is a correcting means for correcting a character in the text buffer 5 by a user operation 7, 9 is a user text created by the user extracted from the text buffer 5, and 10 is a user language model by analyzing the user text 9. It is a user language model learning means.
[0030]
Next, the operation of the speech recognition text input device according to the first embodiment will be described with reference to the drawings. FIG. 2 is a flowchart showing the operation of the user language model learning means of the speech recognition text input device according to Embodiment 1 of the present invention.
[0031]
The voice input means 2 captures the voice 1 spoken by the user. Next, the phonetic character conversion means 4 refers to the acoustic model composed of the phoneme environment-dependent phoneme HMM, and the language models 3 and 3b, and converts the input speech into a word string having the maximum likelihood by a known speech recognition process. Convert.
[0032]
Here, the likelihood of the word string is the product of the likelihood of matching between the word string and the input speech (acoustic likelihood) and the likelihood of the word string based on the language model 3 and the user language model 3b (language likelihood). Calculated as (sum in logarithmic domain).
[0033]
The acoustic likelihood is calculated as the likelihood when the feature vector of the speech 1 is input to the acoustic model in which the phoneme environment-dependent phoneme HMMs are arranged according to the phonetic symbols of the word string. The language model 3 and the user language model 3b are each composed of N-grams of words.
[0034]
For the language likelihood, the likelihood of the word string is calculated with each language model, and the larger likelihood is used.
[0035]
The conversion result from the speech 1 to the word string by the speech character conversion means 4 is stored in the text buffer 5. The display means 6 displays the characters in the text buffer 5 on a computer display.
[0036]
The correction means 8 performs processing such as character insertion / replacement / deletion at the cursor display position indicating the character input position displayed on the computer display by a user operation 7 such as a keyboard input or a mouse, and the like in the text buffer 5. Correct the characters as necessary. Note that the character conversion result itself of the user's utterance can be inserted or replaced at the cursor display position in the same manner as a character string such as a keyboard.
[0037]
The user stores the result of creating the target document by combining the input voice 1 and the user operation 7 as a user text 9 in a file on the computer. At the same time as the user text 9 is stored in the file, the user text 9 is sent to the user language model learning means 10.
[0038]
This user language model learning means 10 includes an analysis dictionary dedicated to text analysis, refers to this dictionary, decomposes the user text 9 into columns of words (called morphemes), and further, the frequency of word chaining N-grams of words are obtained by counting and are output as the user language model 3b.
[0039]
That is, in step 101, the user language model learning means 10 acquires the user text 9 from the file.
[0040]
Next, in step 102, the acquired user text 9 is analyzed and divided into words (also referred to as morphemes) constituting the user text 9. This process is performed using a known morphological analysis technique. The dictionary used for morphological analysis is a dictionary specialized for the target business, and technical terms appearing in the target business are registered.
[0041]
Next, in step 103, a set of consecutive N words (N is 3 for example) is completely extracted from the divided word strings W1, W2, W3,. The N word sets extracted in this way are totaled, and the number of appearances for each N word set is counted.
[0042]
In step 104, the user language model 3b including the number of appearances of N word pairs is output.
[0043]
As described above, the user language model 3b is created from the user text 9, and the phonetic character conversion is performed with reference to the created user language model 3b, so that the user-specific vocabulary and expressions are represented by the user language model 3b. Therefore, it is possible to provide a speech recognition text input device with high recognition performance corresponding to user-specific utterance variations.
[0044]
That is, the speech recognition text input device according to the first embodiment is a speech recognition text input device in which one or more users input speech to create a text, and two or more independent language models 3, 3a. The phonetic character converting means 4 for calculating the likelihood of the word string by referring to any of the two or more independent language models 3 and 3a and converting the speech into characters, and the converted characters Display means 6 for displaying, a correction means 8 for correcting the displayed characters by the user, and a user for updating at least one language model of the two or more language models from the characters corrected by the user The language model learning means 10 is provided.
[0045]
Embodiment 2. FIG.
A speech recognition text input device according to Embodiment 2 of the present invention will be described with reference to the drawings. FIG. 3 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 2 of the present invention.
[0046]
In the first embodiment, after the user text 9 is output to a file, the user language model 3b is created using the user text 9, but in the second embodiment, the user text 9 It is possible to learn at any time during the creation of.
[0047]
In FIG. 3, 7a is a user operation, 8a is a learning start notification means for generating a learning start notification in response to the user operation 7a, and 11 is a user language model based on the learning start notification generated by the learning start notification means 8a. Control means programmed to start processing by the learning means 10. Further, the user language model learning unit 10 is changed so as to input a learning start notification generated by the learning start notification unit 8a. Other configurations are the same as those in the first embodiment.
[0048]
Next, the operation of the speech recognition text input device according to the second embodiment will be described with reference to the drawings. FIG. 4 is a diagram showing a display example by the display means of the speech recognition text input device according to Embodiment 2 of the present invention. FIG. 5 is a flowchart showing the operations of the learning start notifying means, the control means, and the user language model learning means of the speech recognition text input apparatus according to Embodiment 2 of the present invention.
[0049]
The characters converted by the phonetic character conversion means 4 from the voice 1 spoken by the user are temporarily stored in the text buffer 5. The temporarily stored characters are displayed in an editing frame displayed on the computer display by the display means 6, as shown in FIG.
[0050]
Next, the correction means 8 inputs a user operation 7 and edits the text displayed in the editing frame, such as insertion, deletion, and replacement, to create a user text 9.
[0051]
As shown in FIG. 4, the user operation 7 a generates a signal in response to pressing of the learning button displayed near the editing frame of the display unit 6. The learning start notification means 8a generates a learning start notification from the signal generated by the user operation 7a, and the control means 11 starts learning using the user text 9 of the user language model learning means 10.
[0052]
That is, in step 201, the learning start notification means 8a detects that the learning button has been pressed.
[0053]
Next, in step 202, a learning start notification is generated.
[0054]
Next, in step 203, the user language model learning unit 10 activated by the control unit 11 based on the learning start notification acquires the user text 9 from the temporary storage.
[0055]
Next, in step 204, the acquired user text is analyzed and divided into words (also called morphemes) constituting the user text 9. This process is performed using a known morphological analysis technique. The dictionary used for morphological analysis is a dictionary specialized for the target business, and technical terms appearing in the target business are registered.
[0056]
Next, in step 205, all consecutive sets of N words (N is 3 for example) are extracted from the divided word strings W1, W2, W3. The N word sets extracted in this way are totaled, and the number of appearances for each N word set is counted.
[0057]
In step 206, the user language model 3b including the number of appearances of a set of N words is output.
[0058]
As described above, since the user learns at any time in the middle of the text, the text can be input with high accuracy after the learning.
[0059]
That is, in the speech recognition text input device according to the second embodiment, the correction unit 8 is provided with the learning start notification unit 8a that instructs the start of learning after the correction of the character is completed by the user. And a control unit 11 that activates the user language model learning unit 10 after receiving the notification.
[0060]
Embodiment 3 FIG.
A speech recognition text input device according to Embodiment 3 of the present invention will be described with reference to the drawings. FIG. 6 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 3 of the present invention.
[0061]
In the second embodiment, the case where the user language model 3b is learned based on the user operation 7a for learning has been described. However, in the third embodiment, the explicit user operation 7a for learning is not required. It is what.
[0062]
In FIG. 6, 8b is a detection means connected to the correction means 8 to detect the user's text correction. Reference numeral 11 denotes control means programmed to start the learning process of the user language model 3b by the user language model learning means 10 based on the text correction detection by the detection means 8b. Further, the user language model learning unit 10 is changed to input a learning start notification generated by the detection unit 8b.
[0063]
Next, the operation of the speech recognition text input device according to the third embodiment will be described with reference to the drawings. FIG. 7 is a flowchart showing the operations of the detection means, control means, and user language model learning means of the speech recognition text input device according to Embodiment 3 of the present invention.
[0064]
The detection unit 8b generates a correction detection signal and notifies the control unit 11 as soon as the next speech is made by the user and the voice input unit 2 detects the start end of the voice. The control unit 11 activates the user language model learning unit 10, immediately creates the user language model 3b, and the phonetic character conversion unit 4 that refers to the changed user language model 3b uses the recognition result of the input speech as a text buffer. 5 is temporarily stored.
[0065]
Note that the correction detection method may not be detected by the next utterance, but may be detected, for example, when a certain period of time has elapsed since the user last corrected a character in the text buffer 5. Good.
[0066]
That is, in step 301, the detection means 8b detects that the user's next utterance has been made.
[0067]
Next, in step 302, a learning start notification is generated.
[0068]
Next, in step 303, the user language model learning means 10 activated by the control means 11 based on the learning start notification acquires the user text 9 of the previous utterance from the temporary storage.
[0069]
Next, in step 304, the acquired user text 9 is analyzed and divided into words (also referred to as morphemes) constituting the user text 9. This process is performed using a known morphological analysis technique. The dictionary used for morphological analysis is a dictionary specialized for the target business, and technical terms appearing in the target business are registered.
[0070]
Next, in step 305, all sets of consecutive N words (N is 3 for example) are extracted from the divided word strings W1, W2, W3. The N word sets extracted in this way are totaled, and the number of appearances for each N word set is counted.
[0071]
In step 306, the user language model 3b including the number of appearances of N word pairs is output.
[0072]
As described above, since the device detects character correction, it is not necessary to operate the user's learning button.
[0073]
That is, in the speech recognition text input device according to the third embodiment, the correction unit 8 is provided with a detection unit 8b that detects that the character has been corrected by the user, and the user language is detected after the character correction is detected. The control means 11 which starts the model learning means 10 is provided.
[0074]
Embodiment 4 FIG.
A speech recognition text input device according to Embodiment 4 of the present invention will be described with reference to the drawings. FIG. 8 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 4 of the present invention.
[0075]
In the fourth embodiment, in place of the language model 3 in the first to third embodiments, the target business language model 3a learned from the business sentence example to be applied is used.
[0076]
In FIG. 8, 3a is a target business language model.
[0077]
Next, the operation of the speech recognition text input device according to the fourth embodiment will be described with reference to the drawings. FIG. 9 is a flowchart showing the operation of the language model learning means of the speech recognition text input device according to Embodiment 4 of the present invention.
[0078]
The target business language model 3a is learned from the stored sentence examples of the target business. Therefore, the word string similar to the stored sentence example can be recognized by the phonetic character conversion means 4 to refer to the word chain information of the target business language model 3a.
[0079]
That is, in step 401, sentence examples of the target business are manually collected in advance. These sentence examples are assumed to be digitized and stored in a file.
[0080]
Next, in step 402, the language model learning means (not shown) analyzes the sentence example text digitized and stored in the file, and divides it into words (also called morphemes) constituting the sentence example text. This process is performed using a known morphological analysis technique.
[0081]
Next, in step 403, a set of consecutive N words (N is 3 for example) is completely extracted from the divided word strings W1, W2, W3. The N word sets extracted in this way are totaled, and the number of appearances for each N word set is counted.
[0082]
In step 404, the target business language model 3a consisting of the number of appearances of the set of N words is output.
[0083]
As described above, since the target business language model 3a learned from the stored sentence examples of the target business is provided, the sentence examples of the target business can be recognized.
[0084]
That is, in the speech recognition text input device according to the fourth embodiment, at least one of the two or more language models includes the target business language model 3a learned from the sentence examples of the target business.
[0085]
Embodiment 5. FIG.
A speech recognition text input device according to Embodiment 5 of the present invention will be described with reference to the drawings. FIG. 10 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 5 of the present invention.
[0086]
In the fourth embodiment, sentence examples having a high probability of appearing in the professional business sentence are mainly inputted by voice, but in this fifth embodiment, the general sentence that always appears during the creation of the professional business sentence is used. In contrast, even if the recognition rate does not decrease, the recognition rate does not decrease.
[0087]
In FIG. 10, 3a is a target business language model, 3b is a user language model, and 3c is a background language model learned from a general sentence example. The background language model 3c is created by a language model learning unit (not shown).
[0088]
Next, the operation of the speech recognition text input device according to the fifth embodiment will be described with reference to the drawings. FIG. 11 is a flowchart showing the operation of the speech character converting means of the speech recognition text input device according to Embodiment 5 of the present invention.
[0089]
For example, when the user's voice 1 is an intra-business utterance such as “insertion of 6 mg of anticonvulsant determined to thermal convulsions”, the phonetic character conversion unit 4 has a high language likelihood of the target business language model 3a. Outputs the recognition result “6 mg of anticonvulsant determined to be heat convulsions”.
[0090]
In addition, even when the input voice 1 is an out-of-work utterance such as “It seems to have fallen in kindergarten during lunch”, the recognition result “It seems to have fallen in kindergarten in the middle of lunch” is obtained. Output.
[0091]
That is, in step 501, the phonetic character conversion unit 4 inputs the user voice 1.
[0092]
Next, in step 502, text conversion is performed on the user voice 1 using the target business language model 3a, and likelihood 1 is calculated.
[0093]
Next, in step 503, text conversion is performed on the user voice 1 using the user language model 3b, and likelihood 2 is calculated.
[0094]
Next, in step 504, the user speech 1 is subjected to text conversion using the background language model 3c, and likelihood 3 is calculated.
[0095]
In step 505, the text conversion result that gives the maximum likelihood by comparing the likelihood 1, the likelihood 2, and the likelihood 3 is output to the text buffer 5 as the text conversion result.
[0096]
As described above, since the background language model 3c learned from a general sentence example is used, recognition errors can be reduced in sentence input outside the business.
[0097]
That is, in the speech recognition text input device according to the fifth embodiment, the two or more language models are composed of at least three or more language models 3a to 3c, and at least one of the language models is derived from the sentence example of the target business. Learning, and at least one language model uses the language model 3c learned from sentence examples other than the target business.
[0098]
Embodiment 6 FIG.
A speech recognition text input device according to Embodiment 6 of the present invention will be described with reference to the drawings. FIG. 12 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 6 of the present invention.
[0099]
In the fifth embodiment described above, the target business language model 3a is learned in advance. In the sixth embodiment, the target business language model 3a is updated.
[0100]
In FIG. 12, 12 is a sentence example adding means connected to the user text 9, 13 is a sentence example storing means that can be additionally stored by the sentence example adding means 12, and 14 is a target business language model 3 a from the sentence examples stored in the sentence example storing means 13. This is a language model learning means to be created.
[0101]
Next, the operation of the speech recognition text input device according to the sixth embodiment will be described with reference to the drawings. FIG. 13 is a flowchart showing operations of the sentence example adding means and the language model learning means of the speech recognition text input apparatus according to Embodiment 6 of the present invention.
[0102]
First, in step 601, after the user's voice 1 is converted into text (characters) by the voice character converting means 4 and temporarily stored in the text buffer 5, the user text 9 is edited as a result of the user editing by the user operation. Created.
[0103]
Next, in step 602, the sentence example adding means 12 adds the user text 9 to the sentence example storage means 13.
[0104]
In step 603, the language model learning unit 14 creates the target business language model 3 a using the sentence examples stored in the sentence example storage unit 13. Note that the target business language model 3a targets sentence examples of a plurality of users. On the other hand, the user language model 3b is created for each user.
[0105]
As described above, the sentence example adding means 12 stores the user text 9 as a sentence example to learn the target business language model 3a. Therefore, taking into account the user's vocabulary and grammar usage tendency over a relatively long period of time. The user's voice can be recognized.
[0106]
That is, the speech recognition text input device according to the sixth embodiment stores a text example storage unit 13 for storing a text example for use in learning of the target business language model 3a, and stores a text input by the user in the text example storage unit 13. A sentence example adding means 12 for adding the sentence example as a written sentence, and a language model learning means 14 for learning the language model 3a of the target business from the sentence examples stored in the sentence example storage means 13.
[0107]
Embodiment 7 FIG.
A speech recognition text input device according to Embodiment 7 of the present invention will be described with reference to the drawings. FIG. 14 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 7 of the present invention.
[0108]
In the sixth embodiment, the user text 9 is stored in the sentence storage unit 13, but in the seventh embodiment, the user text 9 is analyzed (word division) and stored. is there.
[0109]
In FIG. 14, 12a is a sentence analyzing means for analyzing the user text 9 and outputting an analysis result, 13a is an analyzed sentence example storage means, 12b is an analysis result output from the sentence analysis means 12a is added to the analyzed sentence example storage means 13a. An analyzed sentence example adding means for storing, and 14 is a language model learning means for creating the target business language model 3a using the analyzed sentence examples of the analyzed sentence example storage means 13a.
[0110]
Next, the operation of the speech recognition text input device according to the seventh embodiment will be described with reference to the drawings. FIG. 15 is a flowchart showing operations of the sentence analysis means, the analyzed sentence example addition means, and the language model learning means of the speech recognition text input device according to Embodiment 7 of the present invention.
[0111]
First, in step 701, the user's voice 9 is converted into text (characters) by the voice character converting means 4 and temporarily stored in the text buffer 5, and then the user text 9 is edited as a result of the user's editing by the user operation. Created.
[0112]
Next, in step 702, the sentence analysis unit 12a performs morphological analysis (word division) on the user text 9.
[0113]
Next, in step 703, the analyzed sentence example adding unit 12b adds the analyzed sentence example to the analyzed sentence example storage unit 13a.
[0114]
In step 704, the language model learning unit 14 creates the target business language model 3a using the analyzed sentence example stored in the analyzed sentence example storage unit 13a.
[0115]
As described above, the sentence analysis of the user text 9 is performed and the result is used to create the user language model 3b, and additionally stored in the analyzed sentence example storage unit 13a. The amount of calculation can be done only once.
[0116]
That is, in the speech recognition text input device according to the seventh embodiment, the sentence example stored in the analyzed sentence example storage unit 13a is an analyzed sentence example.
[0117]
Embodiment 8 FIG.
A speech recognition text input device according to Embodiment 8 of the present invention will be described with reference to the drawings. FIG. 16 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 8 of the present invention.
[0118]
In the seventh embodiment, an unknown word is not included in the user's text. In the eighth embodiment, the case where an unknown word is included in the user's text is handled.
[0119]
In FIG. 16, 15 is an unknown word extracting means for extracting an unknown word from the user text 9.
[0120]
Next, the operation of the speech recognition text input device according to the eighth embodiment will be described with reference to the drawings. FIG. 17 is a flowchart showing the operation of the speech recognition text input device according to Embodiment 8 of the present invention.
[0121]
The unknown word extraction means 15 estimates the existence period and pronunciation of the unknown word in the user text 9, and when the unknown word exists, requests the user to speak. The syllable recognition result of the user utterance is added to the vocabulary dictionary as a pronunciation.
[0122]
That is, in step 801, after the user's voice 1 is converted into text (characters) by the voice character converting means 4 and temporarily stored in the text buffer 5, the user text 9 is edited as a result of the user's editing by the user operation. Created.
[0123]
Next, in step 802, the sentence analysis unit 12a performs morphological analysis on the user text 9.
[0124]
Next, when an unknown word exists in the morphological analysis result in step 803, the next step 804 is executed. If it does not exist, step 805 is executed.
[0125]
Next, in step 804, the unknown word extraction means 15 displays the unknown word notation to the user and requests the user to speak to obtain the reading of the unknown word. When the user utters, syllable recognition is performed on the voice inputted by the user, and the syllable recognition result is used as the pronunciation of the unknown word.
[0126]
Next, in step 805, the analyzed sentence example adding unit 12b adds the analyzed sentence example to the analyzed sentence example storage unit 13a.
[0127]
In step 806, the language model learning unit 14 creates the target business language model 3a using the analyzed sentence example stored in the analyzed sentence example storage unit 13a.
[0128]
As described above, since unknown words that do not exist in the sentence analysis dictionary are added to the language models 3a and 3b together with the word chain before and after the sentence, even if the user inputs an unknown word using the correcting means 8, An unknown word that does not exist in the analysis dictionary can be input thereafter.
[0129]
That is, the speech recognition text input device according to the eighth embodiment has an unknown word extraction unit 15 that extracts an unknown word from a corrected character, and the language model learning units 10 and 14 include the unknown word extraction unit 15. It learns word chains including unknown words and unknown words extracted from a language model.
[0130]
Embodiment 9 FIG.
A speech recognition text input device according to Embodiment 9 of the present invention will be described with reference to the drawings. FIG. 18 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 9 of the present invention.
[0131]
In each of the above embodiments, an N-gram of a word is used as a language model. However, in Embodiment 9, an N-gram of a class into which words are classified is used.
[0132]
In FIG. 18, 3a is a target business language model composed of a word N-gram and a class N-gram, 41 is a preliminary search means, and 42 is a reference search for the target business language model 3a to calculate the language likelihood of the word string hypothesis in the word string search. The language probability estimation means 43 for estimating the language probability for the search is connected to the language probability estimation means 42. The word probability search means 43 is connected to the language probability estimation means 42.
[0133]
Next, the operation of the speech recognition text input device according to the ninth embodiment will be described with reference to the drawings. FIG. 19 is a flowchart showing the operation of the speech character converting means of the speech recognition text input device according to Embodiment 9 of the present invention.
[0134]
Each of the target business language model 3a and the user language model 3b includes an N-gram of a word and an N-gram of a class into which words are classified (N = 2 and N = 3).
[0135]
The preliminary search unit 41 of the phonetic character conversion unit 4 calculates the acoustic likelihood of the speech 1 using the built-in phoneme environment-dependent phoneme HMM as an acoustic model, and the target business language model 3a and the user language model 3b with N = 2. The language likelihood with the larger language likelihood is calculated for each, and a candidate lattice of a word string having a large product (sum in the logarithmic domain) of the acoustic likelihood and the language likelihood is output.
[0136]
Next, the word string search means 43 drives the language probability estimation means 42 to output a word string that maximizes the likelihood of N-grams with N = 3 from the word string candidate lattice. As the likelihood of the word string used by the word string search means 43, the language probability estimation means 42 hits (exists) the word string portion in the N-gram of the word (the word string sequence exists in the learning sentence example). In that case, the likelihood of the N-gram of the word is used. When the word N-gram is not hit, the N-gram likelihood of the class to which the word belongs is used.
[0137]
That is, in step 901, the preliminary search means 41 creates a word lattice using an acoustic model and a bigram language model with N = 2.
[0138]
Next, in step 902, the word string search means 43 searches the word string from the word lattice using N-grams with N = 3. The probability of N grams is obtained by the language probability estimation means 42.
[0139]
As described above, the N-gram of the word is applied when the word string connection exists in the example sentence, and the N-gram of the class is applied when it does not exist. Likelihood can be given to a word string including a connection, so that it is possible to cope with a case where the number of sentence examples that can be used for learning a language model is small.
[0140]
That is, the speech recognition text input device according to the ninth embodiment includes language probability estimation means 42 that classifies words into classes and estimates the probability of word alignment that is not included in the sentence example from the characteristics of class alignment. The character conversion means 4 uses the probability estimated from the language probability estimation means 42.
[0141]
Embodiment 10 FIG.
A speech recognition text input device according to Embodiment 10 of the present invention will be described with reference to the drawings. FIG. 20 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 10 of the present invention.
[0142]
In the ninth embodiment, the preliminary search means 41 refers to an N-gram language model such as N = 2, but in the tenth embodiment, the language model is not referred to in the preliminary search stage. .
[0143]
In FIG. 20, 41a is a basic symbol string recognizing means, 42 is a language probability estimating means, 44 is a difference model storing the recognition error tendency of the basic symbol string, and 43a is a word string searching means referring to the difference model 44.
[0144]
Next, the operation of the speech recognition text input device according to the tenth embodiment will be described with reference to the drawings. FIG. 21 is a flowchart showing the operation of the speech character converting means of the speech recognition text input device according to Embodiment 10 of the present invention.
[0145]
In the above configuration, the basic symbol string recognition unit 41a incorporates an acoustic model composed of a phoneme environment-dependent phoneme HMM, calculates the acoustic likelihood of the speech 1 for this acoustic model without referring to the language model, and the acoustic likelihood Outputs the basic symbol string that maximizes. Basic symbols are syllables, phonemes, subwords, and the like, and in the tenth embodiment, they are syllables.
[0146]
The word string search means 43a refers to the target business language model 3a and the user language model 3b from the basic symbol string output from the basic symbol string recognition means 41a via the difference model 44 and the language probability estimation means 42, and the likelihood is high. Output the maximum word string.
[0147]
Here, the likelihood of the word string is the likelihood of the difference model 44 in which the basic symbol string recognized by the basic symbol string recognition means 41a appears from the correct basic symbol strings arranged along the phonetic symbol string of the word string. Is calculated as the product (sum in the logarithm region) of the language likelihood of the larger word string by the target business language model 3a and the user language model 3b.
[0148]
That is, in step 1001, the basic symbol string recognition unit 41a creates a basic symbol string for the input speech 1 using an acoustic model.
[0149]
Next, in step 1002, the word string search means 43a searches the word string from the basic symbol string using a language model. The language model is an N-gram model, and the probability of N-gram is obtained by the language probability estimation means 42.
[0150]
As described above, since the basic symbol string recognition unit 41a as the preliminary search unit 41 does not refer to the language model, the recognition process of the basic symbol string is not required for each language model. Can be reduced to once per utterance.
[0151]
That is, in the speech recognition text input device according to the tenth embodiment, the speech character converting unit 4 recognizes the basic symbol sequence 41a without referring to any language model of speech 1 to 2 or more. And word string search means 43a for searching a word string using the two or more language models 3a and 3b from the recognized basic symbol string.
[0152]
Embodiment 11 FIG.
A speech recognition text input device according to Embodiment 11 of the present invention will be described with reference to the drawings. FIG. 22 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 11 of the present invention.
[0153]
In the first embodiment, one user is used at one time, but in this eleventh embodiment, a plurality of users are used at one time.
[0154]
In FIG. 22, reference numeral 200 denotes a network, 201 denotes a host computer, and 202 denotes a client computer. Although not shown, there are a plurality of client computers 202, and a plurality of users share one host computer 201 in the same job.
[0155]
Next, the operation of the speech recognition text input device according to the eleventh embodiment will be described with reference to the drawings. The basic operation is the same as in the tenth embodiment.
[0156]
The host computer 201 incorporates the target business language model 3a, analyzes the user text 9 created by a plurality of users, stores it in the analyzed sentence example storage means 13a, and periodically updates it at regular time intervals, for example, at regular time zones at night. Update them automatically.
[0157]
As described above, the target business language model 3a built in the host computer 201 is shared by a plurality of users and updated based on the user text 9, so that it is possible to cover utterance variations rather than a single user. A wide range of language models can be provided.
[0158]
That is, the speech recognition text input device according to the eleventh embodiment is a system composed of a plurality of client computers 202 connected to the host computer 201, and at least the analyzed sentence example storage means 13a is arranged in the host computer 201. Thus, it is configured to be collectively managed.
[0159]
【The invention's effect】
As described above, the speech recognition text input device according to claim 1 of the present invention includes a language model composed of N-grams of words and a N-gram of words having a vocabulary unique to the user. User-specific vocabulary and expressions can be added from user text created by user operations. A speech recognition process with reference to an acoustic model composed of a phoneme environment-dependent phoneme HMM, the language model, and the user language model. Calculate the likelihood of the word string with the respective language model and based on the larger likelihood, A speech character converting means for converting an input speech into a word string having the maximum likelihood, a text buffer for temporarily storing the word string converted by the speech character converting means, and characters temporarily stored in the text buffer are displayed. A display unit, a correction unit that corrects the characters displayed on the display unit as needed by a user operation to create a user text, and an analysis dictionary dedicated to text analysis that is a dictionary specialized for the target task. The user text is analyzed with reference to the analysis dictionary, divided into words constituting the user text, and a set of consecutive N words (N is an integer) is completely extracted from the divided word sequence. Then, the extracted N word sets are counted to count the number of appearances of each N word set, and the number of appearances of the N word set is added to the user language model to add a user-specific vocabulary, User-specific User language model learning means for learning expressions, so that user-specific vocabulary and expressions can be added to the user language model, and the recognition performance can be enhanced in response to user-specific utterance variations Play.
[0160]
As described above, the speech recognition text input device according to claim 2 of the present invention includes a learning start notification means for generating a learning start notification in response to a user operation, and a learning start notification from the learning start notification means. And a control means for activating the user language model learning means after receiving the text, the text can be learned at any time during the text, and the text can be input with high accuracy after the learning.
[0161]
As described above, the speech recognition text input device according to claim 3 of the present invention, after detecting the character correction by the detection means, the detection means for detecting that the character has been corrected by the correction means, Since it further comprises control means for activating the user language model learning means, there is an effect that the operation of the learning button can be made unnecessary.
[0162]
The speech recognition text input device according to claim 4 of the present invention can recognize the sentence example of the target business because the language model is the target business language model learned from the sentence example of the target business as described above. There is an effect.
[0163]
As described above, since the speech recognition text input device according to claim 5 of the present invention further includes the background language model learned from the general text example, it is possible to reduce recognition errors in text input outside business. Play.
[0164]
As described above, the speech recognition text input device according to claim 6 of the present invention includes a sentence example storage means for storing a sentence example for use in learning of the target business language model, and the sentence example based on the user text. Since it further comprises a sentence example adding means for adding to the storage means and a language model learning means for creating the target business language model using the sentence examples stored in the sentence example storage means, the user's vocabulary over a relatively long period of time In addition, the user's voice can be recognized in consideration of the usage tendency of the grammar.
[0165]
As described above, the speech recognition text input device according to claim 7 of the present invention analyzes the analyzed text example storage means for storing the analyzed text example for use in learning the target business language model, and analyzes the user text. A sentence analyzing means for outputting the analysis result, an analyzed sentence example adding means for additionally storing the analyzed sentence example analyzed by the sentence analyzing means in the analyzed sentence example storing means, and the analyzed sentence example storing means Language model learning means for creating the target business language model using the analyzed sentence example, and the user language model learning means creates the user language model using the analyzed sentence example from the sentence analyzing means. Therefore, there is an effect that the calculation amount required for the analysis of the user text can be completed only once.
[0166]
As described above, the speech recognition text input device according to claim 8 of the present invention, when there is an unknown word in the user text, uses the syllable recognition result of the user utterance as the pronunciation as a pronunciation to the text analysis means. Since it is further provided with an unknown word extraction means to be delivered, there is an effect that an unknown word that does not exist in the sentence analysis dictionary can be subsequently input even if the user inputs an unknown word by the correction means.
[0167]
In the speech recognition text input device according to claim 9 of the present invention, as described above, the speech character conversion unit classifies the words of the language model into classes, and arranges words that are not included in the sentence examples from the characteristics of class arrangement. A language probability estimating means for estimating a probability of the word, a preliminary search means for creating a word lattice using the probability estimated by the language probability estimating means, and the word lattice using the probability estimated by the language probability estimating means Therefore, it is possible to cope with a case where the number of sentence examples that can be used for learning the language model is small.
[0168]
In the speech recognition text input device according to claim 10 of the present invention, as described above, the speech character conversion unit classifies the words of the language model into classes, and arranges words that are not included in the sentence example from the characteristics of class arrangement. Language probability estimation means for estimating the probability of recognition, a difference model storing a recognition error tendency of a basic symbol string, basic symbol string recognition means for recognizing a basic symbol string without referring to the language model, and the recognized Since it has a word string search means for searching a word string using the language model from the basic symbol string via the difference model and the language probability estimation means, the likelihood calculation can be reduced to once per utterance. Play.
[0169]
As described above, the speech recognition text input device according to claim 11 of the present invention is a speech input means for inputting speech, a text buffer for temporarily storing a word string converted from speech to characters, and temporary storage in the text buffer. A client computer having display means for displaying the displayed characters, and correction means for correcting the characters displayed on the display means as needed by a user operation to create user text; and connected to the client computer via a network A language model consisting of N-grams of words, having a user-specific vocabulary and consisting of N-grams of words User-specific vocabulary and expressions can be added from user text created by user operations. A user language model, an acoustic model composed of a phoneme environment-dependent phoneme HMM, the language model, and the user language model, Calculate the likelihood of the word string with the respective language model and based on the larger likelihood, A speech character converting means for converting the input speech into a word string having the maximum likelihood; and an analysis dictionary dedicated to text analysis which is a dictionary specialized for the target business, and the user text is referred to the analysis dictionary. Are divided into words constituting the user text, and all consecutive sets of N words (N is an integer) are extracted from the divided word strings, and the extracted sets of N words are aggregated to obtain N Count the number of occurrences for each set of words, add the number of occurrences of a set of N words to the user language model, User-specific Since the host computer having the user language model learning means for learning the expression is provided, it is possible to provide a language model having a wider coverage with respect to utterance variations than a single user.
[Brief description of the drawings]
FIG. 1 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 1 of the present invention.
FIG. 2 is a flowchart showing the operation of the user language model learning means of the speech recognition text input device according to Embodiment 1 of the present invention.
FIG. 3 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 2 of the present invention.
FIG. 4 is a diagram showing a display example of display means of a speech recognition text input device according to Embodiment 2 of the present invention.
FIG. 5 is a flowchart showing operations of learning start notifying means, control means, and user language model learning means of the speech recognition text input device according to Embodiment 2 of the present invention;
FIG. 6 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 3 of the present invention.
FIG. 7 is a flowchart showing operations of a detection unit, a control unit, and a user language model learning unit of a speech recognition text input device according to Embodiment 3 of the present invention.
FIG. 8 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 4 of the present invention.
FIG. 9 is a flowchart showing the operation of the language model learning means of the speech recognition text input device according to Embodiment 4 of the present invention.
FIG. 10 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 5 of the present invention.
FIG. 11 is a flowchart showing the operation of the speech character conversion means of the speech recognition text input device according to Embodiment 5 of the present invention;
FIG. 12 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 6 of the present invention.
FIG. 13 is a flowchart showing operations of a sentence example adding unit and a language model learning unit of the speech recognition text input device according to Embodiment 6 of the present invention;
FIG. 14 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 7 of the present invention.
FIG. 15 is a flowchart showing operations of a sentence analyzing unit, an analyzed sentence example adding unit, and a language model learning unit of a speech recognition sentence input device according to Embodiment 7 of the present invention;
FIG. 16 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 8 of the present invention.
FIG. 17 is a flowchart showing the operation of the speech recognition text input device according to Embodiment 8 of the present invention;
FIG. 18 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 9 of the present invention.
FIG. 19 is a flowchart showing the operation of the speech recognition text input device according to Embodiment 9 of the present invention.
FIG. 20 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 10 of the present invention.
FIG. 21 is a flowchart showing the operation of the speech recognition text input device according to Embodiment 10 of the present invention.
FIG. 22 is a diagram showing a configuration of a speech recognition text input device according to Embodiment 11 of the present invention.
FIG. 23 is a diagram showing a configuration of a conventional speech recognition text input device.
[Explanation of symbols]
2 voice input means, 3 language model, 3a target business language model, 3b user language model, 3c background language model, 4 speech character conversion means, 5 text buffer, 6 display means, 7 user operation, 7a user operation, 8 correction means 8a learning start notification means, 8b detection means, 9 user text, 10 user language model learning means, 11 control means, 12 sentence example addition means, 12a sentence analysis means, 12b analyzed sentence example addition means, 13 sentence example storage means, 13a analysis Example sentence storage means, 14 language model learning means, 15 unknown word extraction means, 41 preliminary search means, 41a basic symbol string recognition means, 42 language probability estimation means, 43 word string search means, 43a word string search means, 44 differential model , 200 network, 201 host computer, 202 client computer.

Claims (11)

A language model consisting of N-grams of words;
Ri Do from the N-gram of the word has a user-specific vocabulary, and user language model that user-specific vocabulary, Ru can add representation from user text that is created by the user operation,
Voice input means for inputting voice;
With reference to an acoustic model consisting of a phoneme environment-dependent phoneme HMM, the language model, and the user language model, the likelihood of a word string is calculated with the respective language model by speech recognition processing to obtain a larger likelihood. Based on the phonetic character conversion means for converting the input speech into a word string with the maximum likelihood,
A text buffer for temporarily storing the word string converted by the phonetic character conversion means;
Display means for displaying characters temporarily stored in the text buffer;
Correction means for correcting the characters displayed on the display means as needed by user operation to create user text;
It has an analysis dictionary dedicated to text analysis, which is a dictionary specialized for the target business, analyzes the user text with reference to the analysis dictionary, divides it into words constituting the user text, and the divided words All the consecutive N word pairs (N is an integer) are extracted from the sequence of N, the extracted N word pairs are totaled, the number of occurrences for each N word group is counted, and the number of occurrences of the N word group is calculated. A speech recognition sentence input device comprising: a user language model learning means for learning a user- specific vocabulary and a user-specific expression in addition to the user language model. Learning start notification means for generating a learning start notification in response to a user operation;
The speech recognition text input device according to claim 1, further comprising: a control unit that activates the user language model learning unit after receiving a learning start notification from the learning start notification unit. Detecting means for detecting that the character has been corrected by the correcting means;
The speech recognition text input device according to claim 1, further comprising: a control unit that activates the user language model learning unit after the character correction is detected by the detection unit. The speech recognition text input device according to claim 1, wherein the language model is a target business language model learned from a sentence example of the target business. The speech recognition sentence input device according to claim 4, further comprising a background language model learned from a general sentence example. A sentence example storage means for storing a sentence example for use in learning of the target business language model;
Sentence example adding means for adding a sentence example to the sentence example storage means based on the user text;
The speech recognition text input device according to claim 4, further comprising: a language model learning unit that creates the target business language model using a sentence example stored in the sentence example storage unit. An analyzed sentence example storage means for storing an analyzed sentence example for use in learning of the target business language model;
Sentence analysis means for analyzing the user text and outputting an analysis result;
An analyzed sentence example adding means for additionally storing the analyzed sentence example analyzed by the sentence analyzing means in the analyzed sentence example storage means;
Language model learning means for creating the target business language model using the analyzed sentence examples stored in the analyzed sentence example storage means,
The speech recognition text input device according to claim 4, wherein the user language model learning unit creates the user language model by using an analyzed sentence example from the text analysis unit. 8. The method according to claim 7, further comprising: an unknown word extracting unit that, when an unknown word is present in the user text, delivers a syllable recognition result of a user utterance as a pronunciation together with the unknown word to the sentence analyzing unit Voice recognition text input device. The phonetic character conversion means includes:
Language probability estimation means for classifying the words of the language model into classes, and estimating the probability of arranging words not in the sentence example from the characteristics of the class arrangement;
Preliminary search means for creating a word lattice using the probabilities estimated by the language probability estimation means;
The speech recognition according to any one of claims 1 to 8, further comprising: a word string search unit that searches a word string from the word lattice using the probability estimated by the language probability estimation unit. Text input device. The phonetic character conversion means includes:
Language probability estimation means for classifying the words of the language model into classes, and estimating the probability of arranging words not in the sentence example from the characteristics of the class arrangement;
A difference model that stores the recognition error tendency of the basic symbol string,
Basic symbol string recognition means for recognizing a basic symbol string without referring to the language model;
9. A word string search means for searching a word string from the recognized basic symbol string using the language model via the difference model and the language probability estimation means. The speech recognition text input device according to any of the above. Voice input means for inputting voice,
A text buffer that temporarily stores word strings converted from speech to characters,
Display means for displaying characters temporarily stored in the text buffer; and a client computer having correction means for correcting the characters displayed on the display means as needed by user operation to create user text;
Connected to the client computer through a network;
A language model consisting of N-grams of words,
Ri Do from the N-gram of the word has a user-specific vocabulary, user language model from the user text that is created by the user operation user-specific vocabulary, Ru can be added to representation,
With reference to an acoustic model consisting of a phoneme environment-dependent phoneme HMM, the language model, and the user language model, the likelihood of a word string is calculated with the respective language model by speech recognition processing to obtain a larger likelihood. Based speech conversion means for converting the input speech into a word string with the maximum likelihood, and an analysis dictionary dedicated to text analysis, which is a dictionary specialized for the target business, with reference to the analysis dictionary The user text is analyzed, divided into words constituting the user text, a set of consecutive N words (N is an integer) is extracted from the divided word sequence, and the set of extracted N words is tabulated. User language model learning means for counting the number of appearances for each set of N words and adding the number of appearances of the set of N words to the user language model to learn user-specific vocabulary and user-specific expressions A speech recognition text input device, comprising: a host computer having:

Images