JP2006301102A - Voice recognition device and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To obtain a voice recognition device which can be adapted to various systems such as a sound model without preholding information regarding a user. <P>SOLUTION: A screen control description language analysis section 3 analyzes specific information included in a description language for screen control. An adaptation information determination section 9 holds data of adaptation information corresponding to the specific information and determines adaptation information corresponding to the specific information based upon an analysis result of the screen control description language analysis section 3. A voice recognition section 8 recognizes a voice input to a screen displayed, based upon the description language for screen control according to the adaptation information determined by the adaptation information determination section 9. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a voice recognition apparatus and a program for operating a voice input for an input item on a screen.

  In recent years, the number of people with portable terminals has increased rapidly, and the demand for information input methods to terminals other than keyboard input has increased. One method is voice input. In speech recognition, various techniques have been proposed for a method of adapting a system according to a user and a use environment.

  For example, a user is identified using a customer number or a telephone number, and a plurality of acoustic models prepared in advance for each gender, age, and region are changed according to the user (for example, , See Patent Document 1).

JP 2000-347684 A

  However, in the conventional speech recognition apparatus, it is necessary to hold information about the user in advance. Therefore, there is a problem that the acoustic model cannot be changed in a system used by an unspecified number of people.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a speech recognition apparatus that can be applied to various systems such as an acoustic model without previously storing information about a user. And

  The speech recognition apparatus according to the present invention converts the specific information included in the screen control description language into specific information based on the analysis result of the screen control description language analysis unit and the screen control description language analysis unit. An adaptation information determination unit for determining the corresponding adaptation information, and based on the adaptation information determined by the adaptation information determination unit, the speech input to the screen displayed based on the screen control description language And a voice recognition unit that performs voice recognition.

  Since the speech recognition apparatus according to the present invention performs speech recognition based on the adaptation information corresponding to specific information included in the screen control description language, the speech recognition apparatus does not hold information about the user in advance. It is possible to obtain a speech recognition apparatus that can be applied to various systems such as models.

Embodiment 1 FIG.
1 is a block diagram showing a speech recognition apparatus according to Embodiment 1 of the present invention.
In the figure, the speech recognition apparatus includes a screen control description language acquisition unit 1, a screen display unit 2, a screen control description language analysis unit 3, a system adaptation information holding unit 4, a system data holding unit 5, a system adaptation unit 6, It comprises a voice acquisition unit 7 and a voice recognition unit 8.

  The screen control description language acquisition unit 1 is a functional unit that acquires a screen control description language for displaying a plurality of input items on the screen. The screen display unit 2 is a functional unit that performs screen display based on the screen control description language acquired by the screen control description language acquisition unit 1, and includes a control unit for screen control and a display unit such as a display. It is. The screen control description language analysis unit 3 is a functional unit that analyzes specific information included in the screen data based on the screen control description language acquired by the screen control description language acquisition unit 1.

  The system adaptation information holding unit 4 has an adaptation correspondence table showing correspondence between specific information analyzed by the screen control description language analysis unit 3 and a noise model as adaptation information provided in advance. This is a functional unit that selects noise model information corresponding to specific information from the adaptation correspondence table and outputs the selected information to the system data holding unit 5 as adaptation correspondence information. The system data holding unit 5 holds an actual noise model as an adaptation model, and is configured to output a noise model selected based on the adaptation correspondence information from the system adaptation information holding unit 4. The system adaptation unit 6 is a functional unit that determines a noise model corresponding to the screen and outputs this as adaptation information to the speech recognition unit 8. The system adaptation information holding unit 4 to the system adaptation unit 6 constitute an adaptation information determination unit 9 that determines adaptation information corresponding to specific information analyzed by the screen control description language analysis unit 3. is doing.

  The voice acquisition unit 7 is a functional unit that acquires a user's voice, and includes a voice input circuit and the like. The voice recognition unit 8 performs voice recognition based on the noise model as adaptation information from the system adaptation unit 6 and the voice information from the voice acquisition unit 7, and outputs the recognition result to the screen display unit 2. It is.

  In addition, including the following embodiments, the speech recognition apparatus of the present invention is realized by using a computer, and the configuration of the screen control description language acquisition unit 1 to the speech recognition unit 8 (the screen display unit 2 is a control part). Is realized by a program corresponding to each function and hardware such as a CPU and a memory for executing these programs.

Next, the operation of the first embodiment will be described.
In the present embodiment, description will be made as HTML (HyperText Markup Language), which is a markup language supported by a typical Web browser, as a screen control description language. In the present embodiment, a case where voice input is performed on a page having a plurality of input items for facility inspection work will be described as an example. Furthermore, in the present embodiment, a method for selecting and setting a noise model using a URL (Uniform Resource Locator) of an HTML document as specific information included in the screen control description language will be described.

First, an example site will be described.
FIG. 2 shows an HTML document of the inspection page in the facility 1.
Here, the URL of the inspection page in the facility 1 is “http://www.tenken.ne.jp/place1.html”.
FIG. 3 shows the screen display contents of the inspection page in the facility 1. In the inspection page for the facility 1, the temperature at the center of the facility 1 is input by the input item 101, and the temperature at the facility outlet of the facility 1 is input by the input item 102. After inputting the input items 101 and 102 on the inspection page in the facility 1, when the input determination button 103 is pressed, the inspected data is input to the system.

FIG. 4 shows an HTML document of the inspection page in the facility 2.
Here, the URL of the inspection page in the facility 2 is “http://www.tenken.ne.jp/place2.html”. FIG. 5 shows the screen display contents of the inspection page in the facility 2. On the inspection page in the facility 2, the temperature at the center of the facility 2 is input by the input item 201, and the temperature at the facility outlet of the facility 2 is input by the input item 202. After inputting the input items 201 and 202 on the inspection page in the facility 2, when the input determination button 203 is pressed, the inspected data is input to the system.

Here, the information about each facility is shown in FIG.
In the case of the inspection as described above, the noise environment may vary greatly depending on the inspection place as shown in FIG. When speech recognition is performed using the same noise model under such different noise environments, there is a possibility that data mismatch occurs and high recognition performance cannot be obtained. In order to cope with such a problem, in this embodiment, the adaptation information determination unit 9 switches the system data (adaptation information) to be used according to the display screen. That is, the system adaptation information holding unit 4 selects a noise model corresponding to the information analyzed by the screen control description language analysis unit 3, extracts the noise model from the system data holding unit 5, and the system adaptation unit 6 And output as adaptation information.

  Here, the noise model is a collection of noise under a voice recognition usage environment and information about its characteristics. This noise model is information used as a part of the acoustic model for the probability calculation for the silent portion of the input data. In the present embodiment, it is assumed that a predetermined acoustic model is used as the acoustic model.

Next, the operation of this embodiment will be described using a specific example.
Hereinafter, the system adaptation operation when the inspection page of FIG. 2 is displayed on the screen will be described with reference to the flowchart of FIG.

  First, in step ST101, the screen control description language acquisition unit 1 starts from the network access destination whose URL is “http://www.tenken.ne.jp/place1.html”. Get display screen information. Here, the HTML document shown in FIG. 2 describing the display screen configuration is acquired.

In step ST102, based on the HTML document shown in FIG. 2 acquired by the screen control description language acquisition unit 1, the screen display unit 2 displays the page shown in FIG.
In step ST103, page information such as an HTML document acquired by the screen control description language acquisition unit 1 is analyzed by the screen control description language analysis unit 3. As a result of the analysis, the URL “http://www.tenken.ne.jp/place1.html” of the HTML document is obtained.

  The association information between the URL and the noise model held in the system data holding unit 5 is held as an adaptation correspondence table in the system adaptation information holding unit 4 and stored in a format as shown in FIG. 8, for example. Is done. As can be seen from FIG. 8, when the character string “place1” is included in the URL, it is possible to specify that the facility 1 is the place of use of the present system.

  For example, the system data holding unit 5 holds the noise model shown in FIG. 9 as an adaptation model. Here, the noise model for the facility 1 is assumed to be a noise model obtained by learning the environmental noise of the facility 1, that is, noise such as a telephone bell sound or talking voice. Similarly, the noise model for the facility 2 and the noise model for the facility 3 are also noise models obtained by learning the environmental noise of the facility 2 and the facility 3, respectively. The default noise model is a noise model for setting when the analysis result in the screen control description language analysis unit 3 and the information held in the system adaptation information holding unit 4 are not associated with each other. is there.

  In step ST104, the URL obtained as a result of the analysis by the screen control description language analysis unit 3 is collated with the adaptation correspondence table shown in FIG. 8 held in the system adaptation information holding unit 4. As a result, since the character string “place1” is included in the URL, “the noise model for facility 1” is output to the system data holding unit 5 as adaptation correspondence information. As a result, the system data holding unit 5 selects “the noise model for facility 1” as the noise model used in the page shown in FIG. In step ST105, the system adaptation unit 6 switches to the selected noise model.

  As shown in FIG. 10, the set noise model is used in the arithmetic processing for the silent portion of the input data as part of the default acoustic model in the probability calculation processing in the speech recognition unit 8. Note that the speech segment detection processing, acoustic feature extraction processing, probability calculation processing, and recognition candidate search processing using the recognition dictionary in the speech recognition unit 8 are known processing, and thus description thereof is omitted here.

  By such an operation, an appropriate noise model is set even in a noisy environment, so that recognition performance is improved. Also, if the URL of the display screen obtained by analysis by the screen control description language analysis unit 3 is not in the adaptation correspondence table held in the system adaptation information holding unit 4, information indicating this Is output to the system data holding unit 5 as adaptation support information, and the default noise model is output from the system data holding unit 5 to the system adaptation unit 6. In step ST106, the default noise model is applied to the system adaptation as adaptation information. Set in section 6.

  The same process as described above is performed when the screen control description language acquisition unit 1 acquires display screen information from a network access destination whose URL is “http://www.tenken.ne.jp/place2.html”. . Since the character string “place2” is included in the URL of the display screen, the “facility” is used as the noise model used on the page shown in FIG. 5 from the adaptation correspondence table held in the system adaptation information holding unit 4. "Noise model for 2" is selected and output as adaptation correspondence information. Based on this adaptation correspondence information, the system data holding unit 5 selects the "Noise model for facility 2" from the held noise models. The system adaptation unit 6 switches to the selected noise model.

  As described above, by specifying the use location using the URL of the display screen, it is possible to appropriately set the noise model even when the noise environment varies greatly depending on the use location.

  In the description of the present embodiment, only the URL of the HTML document is used as the information acquired by the screen control description language analysis unit 3, but any display screen analysis information can be used. For example, it is possible to use a character string described on the display screen. Further, in describing the present embodiment, an HTML document is used as a screen control description language, but the present invention is not limited to this, and any description language can be used.

  As described above, according to the speech recognition apparatus of the first embodiment, the screen control description language analysis unit that analyzes specific information included in the screen control description language, and the adaptation information corresponding to the specific information are provided. And an adaptation information determination unit for determining adaptation information based on the analysis result in the screen control description language analysis unit, and voice input to the screen displayed based on the screen control description language. Since it has a voice acquisition unit to be acquired and a voice recognition unit that performs voice recognition of the voice acquired by the voice acquisition unit based on the adaptation information determined by the adaptation information determination unit, information on the user is held in advance It is possible to obtain a speech recognition apparatus that can be applied to various systems such as an acoustic model without having to do so.

  Further, according to the speech recognition apparatus of the first embodiment, the adaptation information determination unit determines the noise model corresponding to the display screen as the adaptation information, and the noise environment greatly varies depending on the place of use. Even in this case, it is possible to appropriately set the noise model.

  In addition, according to the control program for the speech recognition apparatus of the first embodiment, the computer includes a screen control description language analysis unit that analyzes specific information included in the screen control description language, and a screen control description language analysis. Based on the analysis result in the unit, an adaptation information determination unit that determines the adaptation information corresponding to the specific information, based on the adaptation information determined by the adaptation information determination unit, based on the screen control description language Since it is made to function as a voice recognition device including a voice recognition unit that performs voice recognition of voice input to the displayed screen, various information such as an acoustic model can be obtained without holding information about the user in advance. It is possible to implement a speech recognition apparatus that enables adaptation on a simple system on a computer.

Embodiment 2. FIG.
In the second embodiment, a noise superimposed acoustic model for each use environment is prepared as an adaptation model, and the noise superimposed acoustic model is switched according to the display screen.

  In the speech recognition apparatus according to the second embodiment, the configuration in the drawing is the same as that in the first embodiment, and thus will be described with reference to FIG. In the second embodiment, the adaptation correspondence table held by the system adaptation information holding unit 4 is configured to indicate the correspondence between the specific information acquired by the screen control description language analysis unit 3 and the noise superimposed acoustic model. Then, when specific information is given, the system adaptation information holding unit 4 is configured to output information on the noise superimposed acoustic model as adaptation correspondence information based on this adaptation correspondence table. Further, the system data holding unit 5 has a noise superimposed acoustic model as an adaptation model. When the adaptation correspondence information is given from the system adaptation information holding unit 4, the noise superimposed acoustic model corresponding thereto is provided. Is selected and output.

FIG. 11 is an explanatory diagram of an adaptation information table held by the system adaptation information holding unit 4.
FIG. 12 is an explanatory diagram of a noise superimposed acoustic model held by the system data holding unit 5.
The noise superimposing acoustic model shown in these drawings describes information on basic sound units (consonants, vowels, etc.) used for speech recognition, learned from speech data on which noise is superimposed in advance.

  The system adaptation unit 6 is configured to output the noise superimposed acoustic model output from the system data holding unit 5 as adaptation information. The other configuration is the same as that of the first embodiment shown in FIG.

  Next, the operation of the second embodiment will be described. It should be noted that the pages and usage environment used for explaining the operation of the present embodiment are the same as those of the first embodiment.

FIG. 13 is a flowchart showing the operation of determining the adaptation model in the second embodiment.
In the present embodiment, as in the first embodiment, an operation of system adaptation when the inspection page in FIG. 2 is displayed on the screen will be described.

  First, in step ST201, the screen control description language acquisition unit 1 acquires display screen information from a network access destination whose URL is “http://www.tenken.ne.jp/place1.html”. Here, the HTML document shown in FIG. 2 describing the display screen configuration is acquired.

  In step ST202, the screen display unit 2 displays the page shown in FIG. 3 based on the HTML document shown in FIG. 2 acquired by the screen control description language acquisition unit 1. In step ST203, page information such as an HTML document acquired by the screen control description language acquisition unit 1 is analyzed by the screen control description language analysis unit 3. As a result of the analysis, the URL “http://www.tenken.ne.jp/place1.html” of the HTML document is obtained. The association information between the URL and the noise superimposed acoustic model held in the system data holding unit 5 is held as adaptation correspondence information in the system adaptation information holding unit 4 as shown in FIG. As can be seen from FIG. 11, when the character string “place1” is included in the URL, it is possible to specify that the facility 1 is the place where the present system is used.

  The system data holding unit 5 holds, for example, a noise superimposed acoustic model shown in FIG. Here, the noise superimposing acoustic model for facility 1 is assumed to be an acoustic model obtained by learning using voice data on which environmental noise of facility 1 is superimposed. Similarly, the noise superimposing acoustic model for facility 2 and the noise superimposing acoustic model for facility 3 are acoustic models obtained by learning using voice data in which environmental noises of facilities 2 and 3 are superimposed. The default noise superimposing acoustic model is a noise superimposing acoustic to be set when the analysis result in the screen control description language analysis unit 3 and the information held in the system adaptation information holding unit 4 are not associated with each other. It is a model.

  In step ST204, the URL obtained as a result of analysis by the screen control description language analysis unit 3 is collated with the adaptation correspondence table shown in FIG. 11 held in the system adaptation information holding unit 4. As a result, since the character string “place1” is included in the URL, “noise superimposing acoustic model for facility 1” is selected as the noise superimposing acoustic model used in the page shown in FIG. 3 and this information is adapted. Output as correspondence information. The system data holding unit 5 selects a corresponding model from the held noise superimposed acoustic models based on the adaptation correspondence information, and outputs the selected model. In step ST205, the system adaptation unit 6 switches to the given noise superimposed acoustic model. The set noise superimposing acoustic model is used for probability calculation processing in the speech recognition unit as shown in FIG. As a result, an appropriate noise-superimposed acoustic model is set even in a noisy environment, thereby improving recognition performance.

  If the URL of the display screen obtained by analysis by the screen control description language analysis unit 3 is not in the adaptation correspondence table shown in FIG. 11 held in the system adaptation information holding unit 4, step ST206 is performed. , The default noise superimposing acoustic model is set by the system adaptation unit 6.

  As described above, by specifying the use location using the URL of the display screen, it is possible to appropriately set the noise superimposed acoustic model even when the noise environment varies greatly depending on the use location.

  In the description of the present embodiment, only the URL of the HTML document is used as information acquired by the screen control description language analysis unit 3, but any display screen analysis information can be used. For example, it is possible to use a character string described on the display screen. Furthermore, in describing the present embodiment, an HTML document is used as the screen control description language, but any description language can be used.

  As described above, according to the speech recognition apparatus of the second embodiment, the adaptation information determination unit determines the noise superimposed acoustic model corresponding to the display screen as the adaptation information. It is possible to appropriately set the noise superimposing acoustic model even when the values are greatly different. That is, there is an effect that the speech recognition performance can be improved even if the noise model alone is not sufficiently adapted to the usage environment.

Embodiment 3 FIG.
In the third embodiment, the presence / absence of noise removal processing is set based on specific information analyzed by the screen control description language analysis unit 3. That is, when environmental noise can be assumed, it may be more effective in improving recognition performance to perform noise removal processing. Therefore, in this embodiment, the presence / absence of noise removal processing is set using information acquired by the screen control description language analysis unit 3.

The configuration of the speech recognition apparatus according to the third embodiment in the drawing is the same as the configuration in FIG. 1 except that the system data holding unit 5 is not provided.
The system adaptation information holding unit 4 according to the third embodiment includes an adaptation correspondence table that indicates the presence or absence of noise removal processing corresponding to the specific information analyzed by the screen control description language analysis unit 3.

FIG. 15 is an explanatory diagram of an adaptation correspondence table held by the system adaptation information holding unit 4.
As shown in the figure, the location of use and the presence / absence of noise removal processing are associated with a URL as specific information. Further, the system adaptation unit 6 of Embodiment 3 is a system adaptation information holding unit 4. Based on the determined information indicating presence / absence of noise removal processing, the presence / absence of noise removal processing is output as adaptation information. Since the configuration other than this is the same as that of the first embodiment, description of other configurations is omitted.

Next, the operation of the third embodiment will be described.
FIG. 16 is a flowchart illustrating the setting operation of the noise removal process according to the third embodiment.
In the present embodiment, as in the first embodiment, the system adaptation operation when the inspection page in FIG. 2 is displayed on the screen will be described.

  First, in step ST301, the screen control description language acquisition unit 1 acquires display screen information from a network access destination whose URL is “http://www.tenken.ne.jp/place1.html”. Here, the HTML document shown in FIG. 2 describing the display screen configuration is acquired.

  Next, in step ST302, the screen display unit 2 displays the page shown in FIG. 3 based on the HTML document shown in FIG. 2 acquired by the screen control description language acquisition unit 1. In step ST303, page information such as an HTML document acquired by the screen control description language acquisition unit 1 is analyzed by the screen control description language analysis unit 3. As a result of the analysis, the URL “http://www.tenken.ne.jp/place1.html” of the HTML document is obtained. The association information between the URL and the noise removal process is held as adaptation correspondence information in the system adaptation information holding unit 4 as shown in FIG. As can be seen from FIG. 15, when the character string “place1” is included in the URL, it is possible to specify that the facility 1 is the place where this system is used. As is apparent from FIGS. 6 and 15, in the present embodiment, noise removal processing is performed when the environmental noise is stationary noise, and noise removal processing is not performed when the environmental noise is non-stationary noise. And

In step ST304, the URL obtained as a result of the analysis by the screen control description language analysis unit 3 is collated with the adaptation correspondence table shown in FIG. 15 held in the system adaptation information holding unit 4. As a result, since the character string “place1” is included in the URL, “None” is selected as the noise removal process on the page shown in FIG. The Thereby, in step ST305, the system adaptation part 6 sets so that a noise removal process may not be performed. The presence / absence of the set noise removal processing is applied before the acoustic feature extraction in the speech recognition unit 8 as shown in FIG. That is, if the noise removal processing is “none”, the acoustic feature extraction processing is performed as it is, and if the noise removal processing is “present”, the acoustic feature processing is performed after performing the predetermined noise removal processing. I do.
With such an operation, the presence / absence of noise removal processing can be set appropriately even in a noisy environment, so that recognition performance is improved.

  If the URL of the display screen obtained by analysis by the screen control description language analysis unit 3 is not in the adaptation correspondence table shown in FIG. 15 held in the system adaptation information holding unit 4, step ST306 is performed. The system adaptation unit 6 sets a predetermined noise removal processing.

  As described above, by specifying the use location using the URL of the display screen, it is possible to appropriately set the presence or absence of noise removal processing even when the noise environment varies greatly depending on the use location.

  In the description of the present embodiment, only the URL of the HTML document is used as information acquired by the screen control description language analysis unit 3, but any display screen analysis information can be used. For example, it is possible to use a character string described on the display screen. Furthermore, in describing the present embodiment, an HTML document is used as the screen control description language, but any description language can be used.

  As described above, according to the speech recognition apparatus of the third embodiment, the adaptation information determination unit determines whether or not noise removal processing according to the display screen is performed as the adaptation information. Even when the environment is greatly different, it is possible to appropriately set the presence or absence of noise removal processing.

Embodiment 4 FIG.
In the fourth embodiment, the acoustic model for each age and gender is changed according to the display screen. That is, since the acoustic characteristics of speech vary greatly depending on the age and gender, there is a possibility that a high effect in improving recognition performance cannot be obtained only by adaptation to the usage environment. In order to cope with this, in the fourth embodiment, when the page is a page whose age and sex are limited by the display screen, the acoustic model is changed for each age and sex according to the display screen.

The configuration of the speech recognition apparatus according to the fourth embodiment on the drawing is the same as the configuration of FIG. 1 and will be described with reference to FIG.
The system adaptation information holding unit 4 according to the fourth embodiment includes an adaptation correspondence table indicating acoustic models for each age and gender corresponding to specific information analyzed by the screen control description language analysis unit 3.
FIG. 18 is an explanatory diagram of an adaptation correspondence table held by the system adaptation information holding unit 4.
As shown in the figure, corresponding to the URL as the specific information, the information on the acoustic model for each location and age and gender is associated with the URL. When a URL as specific information is given, the system adaptation information holding unit 4 having such an adaptation correspondence table selects an acoustic model corresponding to the URL and outputs this as adaptation correspondence information. It is configured as follows.

The system data holding unit 5 holds an acoustic model for each age and sex as an adaptation model.
FIG. 19 is an explanatory diagram of an acoustic model for each age and gender.
Here, it is assumed that the acoustic model for adult men is an acoustic model obtained by learning using audio data of adult men. Similarly, the acoustic model for adult women, the acoustic model for elderly men, and the acoustic model for elderly women are acoustic models obtained by learning using audio data of adult women, elderly men, and elderly women, respectively. Shall. In addition, the acoustic model for each predetermined age and gender is obtained when the analysis result in the screen control description language analysis unit 3 is not associated with the information in the adaptation correspondence table held in the system adaptation information holding unit 4. It is an acoustic model for setting.

  The system data holding unit 5 having such an adaptation model selects an acoustic model corresponding to the adaptation correspondence information when the adaptation correspondence information is given from the system adaptation information holding unit 4, and adapts this. It outputs so that it may output to the system adaptation part 6 as a computerized model.

  The system adaptation unit 6 according to the fourth embodiment is configured to output adaptation information based on the acoustic model information output from the system data holding unit 5. Since the other configuration is the same as that of the first embodiment, the description thereof is omitted here.

Next, the operation of the fourth embodiment will be described.
FIG. 20 is a flowchart illustrating an acoustic model setting operation according to the fourth embodiment.
In the present embodiment, as in the first embodiment, an operation of system adaptation when the inspection page in FIG. 2 is displayed on the screen will be described.

  First, in step ST401, the screen control description language acquisition unit 1 acquires display screen information from a network access destination whose URL is “http://www.tenken.ne.jp/place1.html”. Next, in step ST402, the screen display unit 2 displays the page shown in FIG. 3 based on the HTML document shown in FIG. 2 acquired by the screen control description language acquisition unit 1.

  In step ST403, page information such as an HTML document acquired by the screen control description language acquisition unit 1 is analyzed by the screen control description language analysis unit 3. As a result of the analysis, the URL “http://www.tenken.ne.jp/place1.html” of the HTML document is obtained. The association information between the URL and the information held in the system data holding unit 3 is set in the system adaptation information holding unit 4 as shown in FIG. It is held as information for performing As can be seen from FIG. 18, when the character string “place1” is included in the URL, the facility 1 can be specified as the place of use of the present system.

  In step ST404, the URL obtained as a result of analysis by the screen control description language analysis unit 3 is collated with the adaptation correspondence table held in the system adaptation information holding unit 4. As a result, since the character string “place1” is included in the URL, in step ST405, “acoustic model for adult men” is system data as an acoustic model for each age and gender used in the page shown in FIG. The acoustic model held in the holding unit 5 is selected from the acoustic model, and this is output from the system adaptation unit 6 to the speech recognition unit 8 as adaptation information.

  The set age / gender acoustic model is used for the probability calculation processing in the speech recognition unit 8 as shown in FIG. Thereby, even when the input subject's age and gender are different depending on the display screen, the acoustic model is appropriately set for each age and gender, so that the recognition performance is improved.

  If the URL of the display screen obtained by the analysis by the screen control description language analysis unit 3 is not in the adaptation correspondence table held in the system adaptation information holding unit 4, in step ST406, the default age A gender-specific acoustic model is set by the system adaptation unit 6.

  As described above, by setting the acoustic model for each age and gender using the URL of the display screen, it is possible to select the acoustic model appropriately when the user's age and gender can be limited by the inspection location It becomes.

  In the description of the present embodiment, only the URL of the HTML document is used as the information acquired by the screen control description language analysis unit 3, but arbitrary display screen analysis information can be used. For example, it is possible to use a character string described on the display screen. Furthermore, in describing the present embodiment, an HTML document is used as the screen control description language, but any description language can be used.

  As described above, according to the speech recognition apparatus of the fourth embodiment, the adaptation information determination unit determines the acoustic model for each age and sex according to the display screen as the adaptation information. Even when the input subject's age and gender differ depending on the screen, the sound recognition performance can be improved because the acoustic model is appropriately set for each age and gender.

  As the adaptation models to be held, the noise model in the first embodiment, the noise superimposing acoustic model in the second embodiment, and the acoustic model for each age and gender in the fourth embodiment. All of them may be retained and appropriately selected from these. In this way, a more appropriate adaptation model can be set according to conditions and the like, and speech recognition performance can be further improved.

Embodiment 5. FIG.
In the first to third embodiments, the method for setting the presence / absence of the noise model, the noise superimposing acoustic model, and the noise removal processing on the display screen has been described. Here, if an adaptation model is set in accordance with the detailed position of each facility, higher speech recognition performance can be realized. Therefore, in the fifth embodiment, a noise model and a noise superimposition model are set according to the position in the facility.

FIG. 22 is an explanatory diagram showing the relationship between the position in the facility and the noise environment.
As shown in the figure, the inspection location varies depending on the input items even in the same facility, and the noise environment may vary greatly. When speech recognition is performed using the same system data (noise model, noise superimposing acoustic model and presence / absence of noise removal processing) under such different noise environments, data mismatch occurs and high recognition performance is obtained. There is no possibility.

  In order to deal with such a problem, in the fifth embodiment, the system data to be used is switched for each input item on the display screen. In addition, since the structure on the drawing of the speech recognition apparatus in Embodiment 5 is the same as that of FIG. 1, it demonstrates using FIG.

  The screen control description language analysis unit 3 according to the fifth embodiment is configured to analyze the URL of the screen as specific information and to analyze the characters included in the name attribute representing the position of each item. Further, the system adaptation information holding unit 4 includes a noise model corresponding to characters included in the name attribute, a noise superimposing acoustic model, and an adaptation correspondence table indicating the presence / absence of noise removal processing, and based on this adaptation correspondence table An adaptation model such as a noise model, a noise superimposing acoustic model, or a noise removal process is selected.

FIG. 23 is an explanatory diagram of an adaptation correspondence table held by the system adaptation information holding unit 4.
As illustrated, the noise model, the noise superimposing acoustic model, and the information indicating the presence or absence of the noise removal process are associated with the character included in the URL as the specific information and the character indicating the position.

The system data holding unit 5 holds a noise model and a noise superimposed acoustic model, and outputs a corresponding model to the system adaptation unit 6 based on the adaptation correspondence information given from the system adaptation information holding unit 4. It is configured.
FIG. 24 is an explanatory diagram of an adaptation model held by the system data holding unit 5.
As shown in the figure, the system data holding unit 5 has data of a noise model and a noise superimposed acoustic model corresponding to the position of each facility.

  Further, the system adaptation unit 6 according to the fifth embodiment is configured to determine information on an adaptation model such as a noise model and a noise superimposed acoustic model output from the system data holding unit 5 as adaptation information. Since the other configuration is the same as that of the first embodiment, the description thereof is omitted here.

Next, the operation of the fifth embodiment will be described.
FIG. 25 is a flowchart showing a system setting operation according to the fifth embodiment.
In the present embodiment, as in the first embodiment, the system adaptation operation when the inspection page in FIG. 2 is displayed on the screen will be described.

  First, in step ST501, the screen control description language acquisition unit 1 acquires display screen information from a network access destination whose URL is “http://www.tenken.ne.jp/place1.html”. Next, in step ST502, the screen display unit 2 displays the page shown in FIG. 3 based on the HTML document shown in FIG. 2 acquired by the screen control description language acquisition unit 1.

  In step ST503, page information such as an HTML document acquired by the screen control description language acquisition unit 1 is analyzed by the screen control description language analysis unit 3. As a result of the analysis, the URL of the HTML document and the name attribute for each input item are obtained.

  In step ST504, the system adaptation information holding unit 4 obtains the URL and the name attribute character for each input item obtained as a result of the analysis by the screen control description language analysis unit 3, and the information of the adaptation correspondence table held therein. And match. As a result, since the URL includes the character string “place1” and the name attribute of the input item 101 includes the character string “centertemp”, the input item 101 of the page shown in FIG. “Facility 1—Facilities center noise model” is selected as the noise model to be used for voice input, and “Facility 1—Facility center noise model” is selected as the noise superimposed acoustic model. It is sent to the system data holding unit 5.

  Based on these pieces of information, the system data holding unit 5 selects a corresponding model from the held adaptation models, and outputs this to the system adaptation unit 6. In step ST505, the system adaptation unit 6 switches to the selected noise model and noise superimposed acoustic model. In addition, “None” is selected for the noise removal processing and is set by the system adaptation unit 6.

  Since the name attribute of the input item 102 includes the character string “exittemp”, in step ST504, the system adaptation information holding unit 4 performs voice input to the input item 102 on the page shown in FIG. “Facility 1—Facility exit noise model” is selected as the noise model to be used, and “Facility 1—Facility exit noise superimposed acoustic model” is selected and output as the noise superimposed acoustic model. Select and output the corresponding model from the stored data. In step ST505, the system adaptation unit 6 switches to the selected noise model and noise superimposed acoustic model. In addition, “Yes” is selected for the noise removal processing, and is set by the system adaptation unit 6.

  If the URL or name attribute of the display screen obtained by analysis by the screen control description language analysis unit 3 is not in the adaptation correspondence table held in the system adaptation information holding unit 4, in step ST506. Default system data is set by the system adaptation unit 6.

  Appropriate when the system environment is set using the URL of the display screen and the name attribute of the input item as described above, so that the usage environment differs for each input item on the display screen and the noise environment differs greatly. It is possible to select and set system data.

  In the description of the present embodiment, the URL of the HTML document and the name attribute of the input item are used as information acquired by the screen control description language analysis unit 3, but arbitrary display screen analysis information should be used. Is possible. For example, it is possible to use a character string described on the display screen. Furthermore, in describing the present embodiment, an HTML document is used as the screen control description language, but any description language can be used.

  As described above, according to the speech recognition apparatus of the fifth embodiment, the adaptation information determination unit determines the noise model for each input item in the display screen as the adaptation information. It is possible to select and set a noise model appropriately even when the use environment differs for each input item and the noise environment varies greatly.

  Further, according to the speech recognition apparatus of the fifth embodiment, the adaptation information determination unit determines the noise superimposed acoustic model for each input item in the display screen as the adaptation information. It is possible to appropriately select and set the noise superimposing acoustic model even when the usage environment is different for each item and the noise environment is greatly different.

  Further, according to the speech recognition apparatus of the fifth embodiment, the adaptation information determination unit determines whether or not noise removal processing is performed for each input item in the display screen as the adaptation information. Even if the use environment differs for each input item and the noise environment is greatly different, it is possible to appropriately set the presence or absence of noise removal processing.

Embodiment 6 FIG.
In the sixth embodiment, the acoustic model is changed for each age and sex according to the input items on the display screen. In other words, when the age and gender are limited for each input item on the display screen, by changing the acoustic model for each age and gender for each input item on the display screen, higher recognition performance can be achieved for those users. Obtainable. In the present embodiment, a method for changing the acoustic model for each age and sex using the information acquired by the screen control description language analysis unit 3 will be described.

  The configuration of the speech recognition apparatus according to the sixth embodiment on the drawing is the same as the configuration of FIG. 1 and will be described with reference to FIG. Further, it is assumed that the pages and the use environment used for explaining the operation of the present embodiment are the same as those of the fifth embodiment.

The system adaptation information holding unit 4 according to the sixth embodiment includes the sound for each age and gender corresponding to the characters included in the URL and the characters included in the name attribute as specific information analyzed by the screen control description language analysis unit 3. An adaptation correspondence table showing the model is provided.
FIG. 26 is an explanatory diagram of an adaptation correspondence table held by the system adaptation information holding unit 4.
As shown in the figure, the acoustic model information for each age and gender is associated with the character included in the URL as the specific information and the character included in the name attribute.

  The system data holding unit 5 holds an acoustic model for each age and sex as an adaptation model. Since the acoustic model for each age and gender is the same as the acoustic model in the fourth embodiment shown in FIG. 19, the description is omitted here.

  The system adaptation unit 6 according to the sixth embodiment is configured to determine adaptation information based on the acoustic model information output from the system data holding unit 5. Since the configuration other than this is the same as that of the fifth embodiment, description of the other configuration is omitted.

Next, the operation of the sixth embodiment will be described.
FIG. 27 is a flowchart illustrating an acoustic model setting operation according to the sixth embodiment.
In the present embodiment, as in the first embodiment, an operation of system adaptation when the inspection page in FIG. 2 is displayed on the screen will be described.

  First, in step ST601, the screen control description language acquisition unit 1 acquires display screen information from a network access destination whose URL is “http://www.tenken.ne.jp/place1.html”. Next, in step ST602, the screen display unit 2 displays the page shown in FIG. 3 based on the HTML document shown in FIG. 2 acquired by the screen control description language acquisition unit 1.

  In step ST <b> 603, page information such as an HTML document acquired by the screen control description language acquisition unit 1 is analyzed by the screen control description language analysis unit 3. As a result of the analysis, the URL of the HTML document and the name attribute for each input item are obtained.

  In step ST604, the URL obtained as a result of the analysis by the screen control description language analysis unit 3 and the name attribute for each input item are collated with the adaptation correspondence table held in the system adaptation information holding unit 4. Since the URL includes the character string “place1” and the name attribute of the input item 101 includes the character string “centertemp”, the voice to the input item 101 on the page shown in FIG. “Adult male acoustic model” is selected as a noise model to be used for input, and this adaptation correspondence information is sent to the system data holding unit 5.

  Based on these pieces of information, the system data holding unit 5 selects a corresponding “adult male acoustic model” from the held acoustic models, and outputs this to the system adaptation unit 6. In step ST605, the system adaptation unit 6 switches to the selected adult male acoustic model.

  Since the name attribute of the input item 102 includes the character string “exittemp”, in step ST604, the system adaptation information holding unit 4 performs voice input to the input item 102 of the page shown in FIG. As an acoustic model to be used, “acoustic model for elderly men” is selected and output, and the system data holding unit 5 selects and outputs a corresponding model from the held data. In step ST605, the system adaptation unit 6 switches to the selected acoustic model.

  As described above, by setting the acoustic model for each age and gender using the URL of the display screen and the name attribute of the input item, the acoustic model is appropriately used when the user's age and gender can be limited depending on the inspection location. Can be selected.

  In the description of the present embodiment, the URL of the HTML document and the name attribute of the input item are used as information acquired by the screen control description language analysis unit 3, but arbitrary display screen analysis information is used. Is possible. For example, it is possible to use a character string described on the display screen. Furthermore, in describing the present embodiment, an HTML document is used as the screen control description language, but any description language can be used.

  As described above, according to the speech recognition apparatus of the sixth embodiment, the adaptation information determination unit determines the acoustic model for each age and sex for each input item on the display screen as the adaptation information. Therefore, even when the user's age and gender are different for each input item on the display screen, it is possible to appropriately select and set the acoustic model.

  Each embodiment can be implemented by causing a computer to read and execute a recording medium that records a program that implements the functions of the above-described embodiments. Various recording media such as a ROM, a CD-ROM, a DVD-ROM, a floppy (registered trademark) disk, a hard disk, and a memory card can be used as a recording medium for supplying the program.

It is a block diagram which shows the speech recognition apparatus by Embodiment 1 of this invention. It is explanatory drawing which shows the HTML document of the page for an inspection in the facility. It is explanatory drawing which shows the display content of the page for an inspection in the facility. It is explanatory drawing which shows the HTML document of the page for an inspection in the plant | facility 2. FIG. It is explanatory drawing which shows the display content of the page for an inspection in the plant | facility 2. FIG. 3 is an explanatory diagram showing information on a facility to be inspected in Embodiment 1. FIG. 3 is a flowchart of a noise model setting operation in the first embodiment. 6 is an explanatory diagram illustrating an adaptation comparison table according to Embodiment 1. FIG. 6 is an explanatory diagram illustrating an adaptation model according to Embodiment 1. FIG. 6 is an explanatory diagram illustrating adaptation information setting processing in Embodiment 1. FIG. FIG. 11 is an explanatory diagram showing an adaptation comparison table in the second embodiment. 6 is an explanatory diagram illustrating a noise superimposing acoustic model according to Embodiment 2. FIG. 10 is a flowchart of a noise superimposing acoustic model setting operation in the second embodiment. FIG. 10 is an explanatory diagram illustrating adaptation information setting processing in the second embodiment. 10 is an explanatory diagram showing an adaptation correspondence table in Embodiment 3. FIG. 10 is a flowchart illustrating an operation when setting whether or not noise removal processing is performed in the third embodiment. FIG. 10 is an explanatory diagram showing adaptation information setting processing in the third embodiment. 10 is an explanatory diagram showing an adaptation comparison table in Embodiment 4. FIG. FIG. 10 is an explanatory diagram showing an adaptation model in the fourth embodiment. 12 is a flowchart of an acoustic model setting operation for each age and gender in the fourth embodiment. FIG. 10 is an explanatory diagram illustrating adaptation information setting processing in the fourth embodiment. It is explanatory drawing which shows the relationship between the position in the facility in Embodiment 5, and a noise environment. FIG. 10 is an explanatory diagram showing an adaptation correspondence table in the fifth embodiment. FIG. 10 is an explanatory diagram showing an adaptation model in the fifth embodiment. 10 is a flowchart showing a system setting operation in the fifth embodiment. FIG. 20 is an explanatory diagram showing an adaptation correspondence table in the sixth embodiment. 18 is a flowchart of acoustic model setting processing for each age / sex according to the sixth embodiment.

Explanation of symbols

2 screen display unit, 3 screen control description language analysis unit, 7 speech acquisition unit, 8 speech recognition unit, 9 adaptation information determination unit.

Claims (11)

A screen control description language analysis unit for analyzing specific information included in the screen control description language;
An adaptation information determination unit for determining adaptation information corresponding to the specific information analyzed by the description language analysis unit for screen control;
A voice acquisition unit that acquires a voice input to a screen displayed based on the screen control description language;
A speech recognition apparatus comprising: a speech recognition unit that performs speech recognition of the speech acquired by the speech acquisition unit based on the adaptation information determined by the adaptation information determination unit.   The speech recognition apparatus according to claim 1, wherein the adaptation information includes at least one of a noise model, an acoustic model, and a noise superimposed acoustic model.   The speech recognition apparatus according to claim 1, wherein the adaptation information determination unit determines a noise model corresponding to the display screen as the adaptation information.   The speech recognition apparatus according to claim 1, wherein the adaptation information determination unit determines a noise superimposed acoustic model corresponding to the display screen as the adaptation information.   The speech recognition apparatus according to claim 1, wherein the adaptation information determination unit determines the presence or absence of noise removal processing according to the display screen as the adaptation information.   The speech recognition apparatus according to claim 1, wherein the adaptation information determination unit determines an acoustic model for each age and sex according to the display screen as the adaptation information.   The speech recognition apparatus according to claim 1, wherein the adaptation information determination unit determines a noise model for each input item in the display screen as adaptation information.   The speech recognition apparatus according to claim 1, wherein the adaptation information determination unit determines a noise superimposed acoustic model for each input item in the display screen as adaptation information.   The speech recognition apparatus according to claim 1, wherein the adaptation information determination unit determines whether or not noise removal processing is performed for each input item in the display screen as the adaptation information.   The speech recognition apparatus according to claim 1, wherein the adaptation information determination unit determines an acoustic model for each age and each gender for each input item in the display screen as the adaptation information. Computer
A screen control description language analysis unit for analyzing specific information included in the screen control description language;
An adaptation information determination unit that determines adaptation information corresponding to the specific information based on an analysis result in the description language analysis unit for screen control;
A speech recognition unit comprising: a speech recognition unit that performs speech recognition of speech input to a screen displayed based on the screen control description language based on the adaptation information determined by the adaptation information determination unit A program for functioning as a device.

Images