JP2009092693A - Speech recognition device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech recognition device which improves recognition performance of speech and usability. <P>SOLUTION: The speech recognition device A for controlling operation of a control object based on speech uttered by a user, comprises: a speech input section 11 to which speech of the user is input; a storage section 12 in which a recognition object vocabulary related to each operation of the control object is respectively stored in a plurality of uttering styles; a selection section 13 for selecting any uttering style from the plurality of uttering styles; a collating section 14 in which the speech input to the speech input section 11 is collated with the recognition object vocabulary of the uttering style which is selected by the selection section 13 in the recognition object vocabulary stored in the storage section 12, and the recognition object vocabulary is specified based on the collation result; and a control section 16 for operating the control object based on the recognition object vocabulary specified by the collation section 14. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a speech recognition apparatus.

2. Description of the Related Art Conventionally, there is a speech recognition device that controls the operation of a control target based on input speech (see, for example, Patent Document 1).
JP 2004-355854 A

  In the speech recognition apparatus, when only one recognition target vocabulary corresponding to each control content is registered, it is difficult for the speech recognition apparatus to recognize the utterance of a certain user's vocabulary, and the user corresponds to that vocabulary. Control cannot be performed. If the registered recognition target vocabulary is difficult for a user to speak, the usability of the speech recognition apparatus is low for the user.

  For example, as a recognition target vocabulary for operating a certain device “heating device”, the name of the device to be operated “heating”, a request-like word to order the operation of the device “turn on heating”, a command-like word “heating” Vocabulary with various utterance styles can be considered, but it is considered that the vocabulary that is easy to use for each user is different, so narrowing the utterance style of the vocabulary to be recognized to any one in advance is from the viewpoint of usability Even if it sees, it is not preferable.

  However, if the recognition target vocabulary corresponding to each control content is simply registered in a plurality of utterance styles, the speech recognition performance deteriorates as the recognition target vocabulary increases, and the device itself becomes difficult to use.

  The present invention has been made in view of the above reasons, and an object of the present invention is to provide a speech recognition apparatus with improved speech recognition performance and usability.

  According to the first aspect of the present invention, in a voice recognition device that controls the operation of a control target based on a voice uttered from a user, the voice input unit to which the user's voice is input is associated with each control target operation. A vocabulary stored in the storage unit, a storage unit that stores each vocabulary in a plurality of utterance styles, a selection unit that selects one of the utterance styles from the plurality of utterance styles, and a voice that is input to the voice input unit A collation unit that collates with the vocabulary of the utterance style selected by the selection unit and identifies the vocabulary based on the collation result, and a control unit that operates the control target based on the vocabulary identified by the collation unit Features.

  According to the present invention, since the vocabulary corresponding to each of the plurality of utterance styles is set for each operation to be controlled, even when speech recognition is difficult with the vocabulary to be recognized of a certain utterance style, By selecting, the voice recognition performance is improved, and the usability of the voice recognition device is improved for the user. That is, the speech recognition performance and usability can be improved. Further, even when the utterance style is increased, an increase in the recognition target vocabulary actually used for the collation process can be suppressed, and a decrease in speech recognition performance can be prevented.

  The invention of claim 2 comprises a presentation means for presenting a plurality of utterance styles of the vocabulary stored in the storage unit according to claim 1, wherein the selection unit is configured to generate a plurality of utterances based on an input from a user. One of the utterance styles is selected from the styles.

  According to the present invention, even when speech recognition is difficult with a recognition target vocabulary of a certain utterance style, the user can select another utterance style and utter the recognition target vocabulary, thereby facilitating speech recognition. Performance is improved. In addition, since the user can select a speech style suitable for him / her, it is easy for the user to speak, and the usability of the speech recognition apparatus is improved for the user.

  The invention of claim 3 is characterized in that, in claim 1, an utterance style classification means for classifying an utterance style of the voice input to the voice input unit into any one of the plurality of utterance styles, and each of the classified utterance styles Frequency calculating means for calculating the appearance frequency, wherein the selection unit selects an utterance style having the highest appearance frequency from a plurality of utterance styles.

  According to the present invention, the speech recognition performance is improved by selecting the utterance style frequently used by the user and specifying the recognition target vocabulary. In addition, since the user can select the utterance style frequently used by the user without making the user aware of it, the usability of the speech recognition apparatus is improved for the user.

  As described above, the present invention has an effect that speech recognition performance and usability can be improved.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

(Embodiment 1)
The speech recognition apparatus A of the present embodiment has a block configuration shown in FIG. 1 and is installed in a bathroom as shown in FIG. 2, and control targets include air conditioning equipment B1 such as heating equipment, main lighting equipment B2, and indirect. There are devices installed in the bathroom such as the lighting device B3, a bathtub device B4 such as a jet bath disposed in the bathtub Y, a bath TV B5, and a hot water supply device (not shown). Furthermore, you may add not only the apparatus in a bathroom but the apparatus outside a bathroom as a control object.

  The voice recognition device A includes a voice input unit 11, a storage unit 12, a selection unit 13, a collation unit 14, a display panel 15, and a control unit 16. The voice input unit 11 is configured by a microphone. For example, it is installed on the wall surface of the bathroom together with the display panel 15 made of a liquid crystal screen.

  The control unit 16 controls the operation of the control target by outputting a control signal to the control target device. Furthermore, the control unit 16 also performs display control of the display panel 15, and displays an operation screen to assist the control of the device to be controlled on the display panel 15. The display screen has the hierarchical structure shown in FIG. 3, and the air conditioner operation screen G1, the lighting device operation screen G2, the bathtub device operation screen G3, the hot water supply device operation screen G4, and the bus TV operation are arranged below the top screen GT. A screen G5 is prepared, and further below the air conditioner operation screen G1, air conditioner operation screens G11, G12,. . . , Lighting device operation screens G21, G22,. . . The bathtub operation screens G31, G32,. . . In the lower layer of the hot water supply device operation screen G4, the hot water supply device operation screens G41, G42,. . . The bus TV operation screens G51, G52,. . . Are prepared for each.

  When the speech recognition apparatus A is activated, the top screen GT is displayed on the display panel 15. Thereafter, the screen is changed according to the voice input from the voice input unit 11. At this time, the control unit 16 outputs the screen state (display screen ID data, display screen state data, etc.) displayed on the display panel 15 to the collation unit 14. The display screen ID data is, for example, ID0 for the top screen GT, ID1 for the air conditioning equipment operation screen G1, ID2 for the lighting equipment operation screen G2, ID3 for the bathtub equipment operation screen G3, and ID4 for the hot water supply equipment operation screen G4. , ID 5 is preset on the bus television operation screen G5, and the air conditioning equipment operation screens G11, G12,. . . Includes ID11, ID12,. . . , Lighting device operation screens G21, G22,. . . ID21, ID22,. . . , Bathtub equipment operation screens G31, G32,. . . , ID31, ID32,. . . , Hot water supply equipment operation screens G41, G42,. . . , ID41, ID42,. . . , Bus TV operation screens G51, G52,. . . ID51, ID52,. . . Is preset.

  First, the voice input unit 11 collects voices uttered from the user and outputs the voices to the collation unit 14 as voice data. The collation unit 14 performs speech recognition on the speech data by using statistical data on correspondence between the feature amount of speech in the speech database stored in the storage unit 12 and phonemes or vocabulary, and data such as recognition target vocabulary data. Processing is performed, the speech data and the recognition target vocabulary stored in the storage unit 12 are collated, and the recognition target vocabulary having the highest similarity to the speech data is specified.

  The recognition target vocabulary is stored in advance in the storage unit 12 in association with the screen displayed on the display panel 15, and the collation unit 14 associates with the currently acquired display screen ID data and display screen state data. The recognized recognition vocabulary is used for collation processing.

  For example, as shown in FIG. 4, “activate the heating device”, “display the heating device”, “1” corresponding to the display screen ID data “1” and the display screen state data (corresponding to the screen description in FIG. 4) of the air conditioning device operation screen G1. Each operation content such as “start ventilation fan” and “transition of display screen to top screen” is prepared, and a recognition target vocabulary is set corresponding to each operation content. In addition, in response to the display screen ID data “2” and the display screen state data on the lighting device operation screen G2, operations of “turn on main illumination”, “turn on indirect illumination”, and “transition display screen to top screen” are performed. Contents are prepared, and a recognition target vocabulary is set corresponding to each operation content.

  Furthermore, the recognition target vocabulary for each operation content is set corresponding to a plurality of utterance styles, and in this embodiment, a name single style, a request style, a command style, etc. are used as the utterance style.

  For example, as shown in FIG. 4, the recognition target vocabulary of the operation content “activate heating device” at the time of air conditioning device operation is “heating” when the utterance style is a single name style, and when the utterance style is a request style “Turn on heating” is set, and if the utterance style is a command style, “Turn on heating” is set. Also, the vocabulary to be recognized for the operation content “Activate Ventilation Fan” is “Ventilation Fan” when the utterance style is the name-only style, “With Ventilation Fan” when the utterance style is the request style, and the utterance style is the command style. In the case of, “Turn on ventilation fan” is set. Similarly, in the case of other operation contents, a recognition target vocabulary is set for each utterance style. In addition, the vocabulary to be recognized for the operation content “transition from display screen to top screen” is “top screen” when the utterance style is the name alone style, and “transition to the top screen” when the utterance style is the request style. When the utterance style is the command style, “transition to the top screen” is set. In other operations as well, recognition target words are set corresponding to a plurality of utterance styles.

  As described above, in the storage unit 12, the recognition target vocabulary for each utterance style is set in one operation content, but the collation unit 14 is one of the plurality of utterance styles selected by the selection unit 13 described later. Only the recognition target vocabulary corresponding to the utterance style is used for collation processing.

  Next, the speech style selection process will be described. In this embodiment, an utterance style selection screen is prepared in advance as a screen to be displayed on the display panel 15, and the user currently selects a recognition target vocabulary corresponding to the operation content “transition of display screen to utterance style selection screen”. When speaking to the voice input unit 11 in the utterance style that has been selected (in this case, the request style), the collation unit 14 uses only the recognition target vocabulary corresponding to the selected utterance style for the collation process. When the recognition target vocabulary having the highest similarity to the voice data (in this case, “transition to the utterance style selection screen”) is specified as a result of the processing, the display on the display panel 15 transitions to the utterance style selection screen. To do. On the utterance style selection screen, a single name style, a request style, a command style, etc. are displayed as selectable utterance styles.

  The user can select a desired utterance style from a plurality of utterance styles such as a name-only style, a request style, and a command style while viewing the utterance style selection screen. However, if the recognition target vocabulary corresponding to the operation content “selecting the name alone style as the utterance style” is uttered toward the voice input unit 11 in the currently selected utterance style (in this case, the request style), the matching unit 14 Then, when the recognition target vocabulary having the highest similarity with respect to the speech data (in this case, “select a single name style”) is specified as a result of the collation processing, the collation unit 14 performs processing as shown in FIG. In addition, the utterance style ID corresponding to the selected utterance style (in this case, the utterance style ID “1” corresponding to the name-only style) is output to the selection unit 13.

  The selection unit 13 recognizes corresponding to the utterance style (in this case, the name single style) corresponding to the received utterance style ID among the recognition target vocabulary stored in the storage unit 12 as the recognition target vocabulary used for the collation processing. Select the target vocabulary.

  Thereafter, the collation unit 14 uses only the recognition target vocabulary corresponding to any utterance style (in this case, the name single style) selected by the selection unit 13 for the collation processing. When the selected utterance style is a name-only style, for example, the recognition target vocabulary corresponding to the operation contents “activate heating device” and “transition display screen to top screen” are “heating” and “top screen”, respectively. Recognized vocabularies of different utterance styles such as “turn on heating”, “turn on heating”, “transition to top screen”, and “transition to top screen” are not used for the collation process.

  In this manner, the collation unit 14 uses only the recognition target vocabulary of the predetermined utterance style narrowed down from the recognition target vocabulary corresponding to all the utterance styles stored in the storage unit 12 to the speech data. Speech recognition processing for identifying the recognition target vocabulary having the highest similarity. Then, when identifying the recognition target vocabulary having the highest similarity to the voice data, the collation unit 14 outputs a control command signal corresponding to the identified recognition target vocabulary to the control unit 16 as shown in FIG. For example, if the identified recognition target vocabulary is “heating”, the activation command signal of the air conditioning device B1 is output, and if the identified recognition target vocabulary is “main lighting”, the activation command signal of the main lighting device B2 is output. To do.

  Based on the control command signal received from the collation unit 14, the control unit 16 transmits a control signal to the air conditioner B1, the main lighting device B2, the indirect lighting device B3, the bathtub device B4, the bus television B5, and the like to perform each operation. While controlling, the screen of the display panel 15 is changed to the screen relevant to each control content. For example, when the air conditioner B1 is activated, the air conditioner operation screen G11 is displayed on the display panel 15, and when the main illumination device B2 is attached, the illumination device operation screen G21 is displayed on the display panel 15. Also at this time, the control unit 16 outputs the screen state (screen ID data of the display screen, display screen state data, etc.) displayed on the display panel 15 to the verification unit 14.

  Therefore, since the recognition target vocabulary corresponding to each of the plurality of utterance styles is set for each operation content to be controlled, the collation unit 14 can perform speech recognition with a recognition target vocabulary of a certain utterance style. When the user selects another utterance style and speaks the recognition target vocabulary, the speech recognition becomes easy and the speech recognition performance is improved. In addition, since the user can select a speech style suitable for him / her, it is easy for the user to speak, and the usability of the speech recognition apparatus is improved for the user. That is, the speech recognition performance and usability can be improved.

  Further, even when the utterance style is increased, an increase in the recognition target vocabulary actually used for the collation process can be suppressed, and a decrease in speech recognition performance can be prevented.

  Note that the voice recognition device A may be installed in a living space other than the bathroom. For example, when the voice recognition device A is installed in a kitchen and a user inputs a voice, the device control in the kitchen or the screen of the display panel 15 can be changed to display a cooking recipe.

  Further, as the utterance style, as shown in FIG. 7, in addition to the single name style, the request style, and the command style, an unsatisfactory style and an associative style are conceivable. In the dissatisfied style, a vocabulary corresponding to the content of the user's dissatisfaction is registered as a recognition target vocabulary, and device control that resolves the dissatisfaction is associated with the recognition target vocabulary. For example, the recognition target vocabulary “cold” may be associated with the operation content “activate heating device” (the heating device is activated in order to eliminate the dissatisfaction of being cold).

  The associative style registers a word that is unexpectedly generated by the user as a recognition target vocabulary, and associates the device control associated with the recognition target vocabulary with the utterance content. For example, the recognition target vocabulary “cold” may be associated with “activate the heating device” (since it is necessary to warm the feeling after it is cold, the heating device is activated). In addition, for example, the operation content “increased illumination intensity” may be associated with the recognition target vocabulary “sleepy” (because it is necessary to wake up feeling after being sleepy, increase the illumination intensity) ).

(Embodiment 2)
The speech recognition apparatus A according to the present embodiment has the block configuration shown in FIG. 8 and includes the frequency calculation unit 17 in the configuration of the first embodiment. The same reference numerals are given to the same configurations as those in the first embodiment. Description is omitted.

  In the present embodiment, the frequency at which each utterance style is used within a predetermined period is calculated for the voice data from the voice input unit 11, and the utterance style is switched based on the calculated frequency. A flowchart is shown in FIG.

  First, when the collation unit 14 receives voice data input to the voice input unit 11 by the user, the collation unit 14 functions as an utterance style classification unit, and the utterance style of the voice data is set to a plurality of preset utterances. The style is classified into one of the styles, and the utterance style ID (see FIG. 5) corresponding to the classified utterance style is output to the frequency calculation section 17, and the frequency calculation section 17 acquires the utterance style ID (S1). Note that the utterance styles are classified by the collation unit 14 by recognizing features (suffixes, intonations, etc.) for each utterance style using a known speech recognition method.

  The frequency calculation unit 17 functions as a frequency calculation unit that calculates the appearance frequency of each classified utterance style, and decrements the utterance style selection count value (initial value 30) of a built-in timing counter (not shown). (S2) Further, the ID acquisition count for each utterance style is counted by a built-in frequency counter (not shown) (S3). The utterance style selection count value indicates the timing of utterance style selection. When the count value becomes zero, the utterance style is selected (S4). If the count value is not zero, the present process is terminated, and S1 is again performed. Start processing from.

  When the count value becomes zero, the frequency calculation unit 17 resets the utterance style selection count value to the initial value 30 (S5), and then calculates the ratio of the ID acquisition count for each utterance style, so that the ID acquisition count is 30 times. It is determined whether or not there is an utterance style that occupies a majority (S6). If there is an utterance style that occupies the majority, the appearance frequency of the utterance style is the highest, and it is considered that the user likes the utterance style, and the utterance style ID of the utterance style is output to the selection unit 13. In addition, a screen generation command for notifying that the utterance style has been switched is output to the control unit 16 via the verification unit 14 (S7), and the control unit 16 is currently displayed on the display panel 15, for example. The speech style switching notification screen H1 is pasted and displayed on the air conditioning equipment operation screen G11 (FIG. 10A) (FIG. 10B).

  The selection unit 13 selects the recognition target vocabulary corresponding to the utterance style set by the received utterance style ID from the recognition target vocabulary stored in the storage unit 12 as the recognition target vocabulary used for the collation processing.

  Thereafter, the collation unit 14 uses only the recognition target vocabulary corresponding to one of the utterance styles selected by the selection unit 13 for the collation processing.

  In S6, if there is no utterance style that occupies the majority, it is considered that the user's favorite utterance style has not been determined, the utterance style is not switched, this processing ends, and the processing starts again from S1. .

  The frequency calculation unit 17 may output the utterance style ID of the utterance style with the largest number of ID acquisitions to the selection unit 13. The initial value of the utterance style selection count value used by the frequency calculation unit 17 may be other than 30.

  Therefore, since the recognition target vocabulary corresponding to each of the plurality of utterance styles is set for each operation content to be controlled, the collation unit 14 selects the utterance style frequently used by the user and specifies the recognition vocabulary. As a result, the voice recognition performance is improved. In addition, since the user can select the utterance style frequently used by the user without making the user aware of it, the usability of the speech recognition apparatus is improved for the user. That is, the speech recognition performance and usability can be improved.

  Further, even when the utterance style is increased, an increase in the recognition target vocabulary actually used for the collation process can be suppressed, and a decrease in speech recognition performance can be prevented.

It is a figure which shows the block configuration of the speech recognition apparatus of Embodiment 1. It is a figure which shows the speech recognition apparatus installed in the bathroom. It is a figure which shows the hierarchical structure of a display screen. It is a figure which shows a response | compatibility with a display screen and recognition object vocabulary. It is a figure which shows a response | compatibility with speech style and speech style ID. It is a figure which shows a response | compatibility with a recognition object vocabulary and a control command. It is a figure which shows the example of an utterance style. It is a figure which shows the block configuration of the speech recognition apparatus of Embodiment 2. It is a figure which shows the operation | movement flowchart at the time of speech style setting. (A) (b) It is a figure which shows pasting of the speech style switching alert | report screen.

Explanation of symbols

A voice recognition device 11 voice input unit 12 storage unit 13 selection unit 14 collation unit 15 display panel 16 control unit B1 air conditioning device B2 main lighting device B3 indirect lighting device B4 bathtub device B5 bus TV

Claims (3)

In a voice recognition device that controls the operation of a control target based on voice emitted from a user,
A voice input unit for inputting a user's voice;
A storage unit storing vocabulary associated with each action to be controlled in a plurality of utterance styles,
A selection unit for selecting one of the utterance styles from a plurality of utterance styles;
A collation unit that collates speech input to the speech input unit with the vocabulary of the utterance style selected by the selection unit from the vocabulary stored in the storage unit, and identifies the vocabulary based on the collation result;
A speech recognition device comprising: a control unit that operates a control target based on the vocabulary specified by the collation unit.   Presenting means for presenting a plurality of utterance styles of the vocabulary stored in the storage unit, wherein the selection unit selects one of the utterance styles from the plurality of utterance styles based on an input from the user. The speech recognition apparatus according to claim 1, wherein:   Utterance style classification means for classifying the utterance style of the voice input to the voice input unit into any of the plurality of utterance styles; and a frequency calculation means for calculating the appearance frequency of the classified utterance styles, The speech recognition apparatus according to claim 1, wherein the selection unit selects an utterance style having the highest appearance frequency from a plurality of utterance styles.

Images