JP2005227673A - Voice recognition device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice recognition device which can easily be operated almost without depending on visual sense. <P>SOLUTION: When a denial evaluation is inputted in the voice recognition device, information is made for a phoneme text of a next recognition candidate and the device urges an evaluation input of the text. Moreover, an input operation of an affirmative evaluation is made by stroking a touch pad 21 and an input operation of a denial result is made by patting the touch pad 21 so that evaluation inputs are made by the direct feeling operations. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a speech recognition apparatus that recognizes phoneme text from a speech signal.

2. Description of the Related Art Conventionally, a speech recognition device that recognizes phoneme text (uttered character string) from a speech signal is known. In this type of speech recognition apparatus, a plurality of candidate phoneme texts are recognized from the user's voice, and the user operates a predetermined input device to select a correct recognition phoneme text, and feedback from the user Based on the above, there is one that improves the subsequent recognition rate (for example, Patent Document 1).
JP-A-8-202388

  By the way, in recent years, this type of voice recognition device is configured as an in-vehicle device such as a car navigation device, and various voice operations are available.

  However, in the conventional speech recognition apparatus, a plurality of operation buttons are arranged at different places, and in order for the user to determine an appropriate phoneme text, an operation of selecting one phoneme text from a plurality of candidates ( A selection operation) and an operation (decision operation) for confirming the selected phoneme text are required, and an operation that relies on vision is required, which is not preferable as an operation of the in-vehicle device.

  The present invention has been made in view of the above-described circumstances, and an object of the present invention is to provide a voice recognition device that can be easily operated with little dependence on vision.

  In order to solve the above-described problem, the present invention provides a speech recognition unit that recognizes a plurality of candidate phoneme texts from a speech signal, a notification unit that notifies one phoneme text of the plurality of candidates recognized by the speech recognition unit, An input device for inputting a positive or negative evaluation of the phoneme text notified by the notification means, and the notification means inputs an evaluation of another phoneme text when a negative evaluation is input via the input device. The speech recognition apparatus is characterized by informing other phoneme texts of the plurality of candidate phoneme texts. According to this configuration, when a negative evaluation is input, the other phoneme text is notified in order to prompt the input of the evaluation of the other phoneme text, so that it is not necessary to perform an operation for selecting the phoneme text. In this configuration, it is preferable that the input device can input the positive evaluation and the negative evaluation by different operation modes.

  The present invention also provides speech recognition means for recognizing phoneme text from a speech signal, notification means for notifying the phoneme text recognized by the speech recognition means, and positive or negative evaluation of the phoneme text notified by the notification means. An input device for inputting is provided, and the speech recognition device is characterized in that the input device can input the positive evaluation and the negative evaluation by different operation modes. According to this configuration, since the positive evaluation and the negative evaluation can be input by different operation modes, the evaluation can be easily input. In this configuration, it is preferable that the positive evaluation operation mode is a stroke operation. In this case, a swingable operation body is provided, and a stroke operation is detected by detecting the swing of the operation body. preferable.

  Further, in this configuration, it is preferable that the operation mode of the negative evaluation is an operation of tapping. In this case, an operation body provided so as to be movable up and down is provided, and an operation of detecting and tapping the operation body is detected. It is preferable to detect. Further, in each of the above configurations, a sound emitting unit that emits a sound according to the evaluation input through the input device, or a process execution unit that executes a process according to the phoneme text when a positive evaluation is input. It may be provided.

  Further, in this configuration, a storage unit that stores a plurality of recognition algorithms and a plurality of recognition parameters is provided, and the speech recognition unit performs speech based on any of the recognition algorithms and recognition parameters stored in the storage unit. Recognition algorithm with high correlation and high positive evaluation based on the combination of recognition algorithm and recognition parameters used for speech recognition and the frequency of positive evaluation and negative evaluation input at the time of this combination The recognition parameter combination may be determined, and thereafter, speech recognition may be performed using the determined combination.

  In the present invention, the user does not need to perform an operation (selection operation or determination operation) for selecting a phoneme text from among a plurality of recognition candidates, and only needs to perform an evaluation input operation. In addition, since the input device can input positive evaluation and negative evaluation in different operation modes, an operation that does not depend on vision is possible.

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

  FIG. 1 is a block diagram showing an outline of an in-vehicle voice recognition device 10 according to the present embodiment. The voice recognition device 10 is configured as a vehicle-mounted device such as a car navigation device or a car audio device. In addition, in this embodiment, since it is the structure substantially the same as the conventional vehicle-mounted apparatus except the structure regarding speech recognition, only the structure regarding speech recognition is demonstrated.

  The control unit (processing execution means) 11 functions as a speech recognition engine that performs speech recognition processing for recognizing phoneme text from speech signals input via the microphone 12 and controls the entire speech recognition apparatus 10. . An algorithm database (hereinafter referred to as “DB”) 13 is a database that stores a plurality of recognition algorithms, and a recognition parameter DB 14 includes a plurality of recognition parameters (gender-specific, age-specific, and region-specific). This is a database for storing acoustic parameters and the like. Here, the recognition algorithm may be, for example, a DP matching method that normalizes expansion / contraction of the utterance time length using dynamic programming, an HMN statistics establishment method using an HMN (Hidden Markov Model), or an n-Gram method (n An algorithm such as a method using statistical probability between words is employed.

  That is, the control unit 11 can perform speech recognition using a plurality of recognition methods by using a plurality of recognition algorithms stored in the algorithm DB 13 and a recognition parameter stored in the recognition parameter DB 14. .

  The condition storage memory 15 is a non-volatile memory for storing speech recognition conditions. The memory 15 further stores learning data and a statistical language model (for a sequence of N words in which appearance probabilities are consecutive). Data of a model for estimating from the appearance probability is also stored.

  Under the control of the control unit 11, the voice synthesizer 16 performs a voice synthesis process for generating a voice signal of the synthesized voice, and outputs the voice signal to the speaker 17 to emit the synthesized voice. Here, as the speaker 17, a speaker that is preliminarily installed in a vehicle on which the voice recognition device 10 is mounted may be used.

  The user evaluation input device 20 is an input device for inputting a voice recognition evaluation from a user. Hereinafter, the configuration of the user evaluation input device will be described in detail.

  2 is a top view of the user evaluation input device 20, and FIG. 3 is a side sectional view. The user evaluation input device 20 includes a touch pad (operation body) 21 operated by a user, an outer case 24 including a support base 22 that supports the touch pad 21 and a wall portion 23 that surrounds the touch pad 21. And is attached to an apparatus main body (not shown) of the speech recognition apparatus 10.

  More specifically, the touch pad 21 is formed of a resin such as plastic and has a pad portion 21A formed with a gently curved upper surface, and a spring insertion portion 21B provided integrally below the pad portion 21A. A coil spring 25 made of a conductive material such as metal is inserted into the spring insertion portion 21B, and the other end of the coil spring 25 is a convex portion provided on the upper surface of the support base 22. Thus, the touch pad 21 is supported by the support base 22 so as to be swingable and vertically movable.

  The wall portion 23 is formed in a cylindrical shape having an inner diameter larger than the outer diameter of the pad portion 21A, and has a support base so as to have a gap between the pad portion 21A of the touch pad 21 as shown in FIGS. 22 is fixed.

  Further, the touch pad 21 is provided with an electrode 30A around the pad portion 21A, and the electrode 30A is electrically connected to the coil spring 24. An electrode 30B is also provided on the lower surface of the spring insertion portion 21B. Further, the wall portion 23 is also provided with an electrode 30C at a position facing the electrode 30A of the pad portion 21A described above, and the support base 22 is also provided with an upper surface of the convex portion 22A (that is, a position facing the electrode 30B). An electrode 30D is provided on the substrate. Further, the support base 22 is provided with three metal terminals (conduction terminals) 40a, 40b, and 40c. The electrode 30C, the coil spring 24, and the electrode 30D are electrically connected to the metal terminals 40a, 40b, and 40c, respectively. Connected. The three metal terminals 40a, 40b, and 40c are configured such that the continuity between the two terminals is detected by the control unit 11 of the speech recognition apparatus 10.

  With the above configuration, when the touch pad 21 is swung, the electrode 30A of the pad portion 21A and the electrode 30C of the wall portion 23 are in contact with each other, and the metal terminal 40a and the metal terminal 40b are electrically connected via the coil spring 24. (I.e., it functions as a swinging operation (swinging operation) detecting means). In other words, as shown in FIG. 4, when the user performs an operation that simulates the action of stroking a person on the touch pad 21, the metal terminal 40 a and the metal terminal 40 b become conductive. In the present embodiment, this operation is assigned to an affirmative evaluation input operation of “appropriate recognition”, and the control unit 11 detects the continuity between the metal terminal 40a and the metal terminal 40b, thereby allowing the user to It is detected that a positive evaluation is input from.

  When the touch pad 21 is pressed, the electrode 30B of the touch pad 21 and the electrode 30D of the support base 22 come into contact with each other, and the metal terminal 40b and the metal terminal 40c are electrically connected (that is, up and down movement (striking operation)). Functions as a detection means). In other words, as shown in FIG. 5, when the user performs an operation simulating the operation of hitting a person on the touch pad 21, the metal terminal 40a and the metal terminal 40b become conductive. In the present embodiment, this operation is assigned to an input operation for negative evaluation of “inappropriate recognition (incorrect recognition)”, and the control unit 11 conducts the metal terminal 40a and the metal terminal 40b. , It is detected that a negative evaluation is input from the user.

  Next, the voice recognition procedure of the voice recognition apparatus 10 will be described. FIG. 6 is a flowchart showing a voice recognition procedure. When the power is turned on, first, the control unit 11 sets initial conditions (step S1). The initial condition is set by reading the learning data stored in the condition storage memory 15 and selecting a recognition algorithm set as a default from a plurality of recognition algorithms registered in the algorithm DB 13 and in the recognition parameter DB 14. This is a process of selecting a parameter set as a default from a plurality of registered recognition parameters.

  Next, the control unit 11 performs a voice recognition process on condition that voice is input via the microphone 12 (step S2). Since the operation of the voice recognition process is substantially the same as the conventional one, in brief explanation, the control unit 11 sets the recognition parameter selected in the process of step S1 above for the voice signal input through the microphone 12. Using the selected recognition algorithm, and if there is read learning data, the learning data is also used to obtain a plurality of phoneme texts by calculation, and these phoneme texts are rearranged in order of priority and temporarily stored. Store in a storage memory (not shown).

  And the control part 11 performs the alerting | reporting process which alert | reports to a user the phoneme text with the highest possibility (phoneme text with the highest priority) from the phoneme text of a recognition candidate (step S3). Specifically, the control unit 11 performs a process of causing the speech synthesizer 16 to emit a synthesized speech corresponding to the phoneme text from the speaker 17. Thereby, a user (for example, a driver) can recognize a voice recognition result without depending on vision. Since this notification process also prompts the user to input an evaluation as to whether or not the speech recognition result is appropriate, the sound emission content is, for example, “XX (phoneme text as a speech recognition result). ) "Can be used (inquiry format). In the case where the voice recognition device 10 includes a display device such as a liquid crystal display device, the voice recognition result is combined with the sound emission processing or instead of the sound emission processing at the time of the notification processing. May be displayed as text or images.

  After this notification process, the control unit 11 enters an input waiting state for an evaluation on the voice recognition result from the user. When the evaluation is input via the input device 1 (step S4), the process proceeds to the process of step S5.

  Here, as described above, since the input operation for affirmative evaluation is an operation for stroking the touch pad 21 as shown in FIG. 4, it is easy and positive that the user can stroke the touch pad 21 from the viewpoint of the user. It is possible to input an affirmative evaluation by a natural operation that a person can take. Further, since the negative evaluation input operation is also an operation of hitting the touch pad 21 as shown in FIG. 5, it is easy for the user to hit the touch pad 21 and can be taken by a person in the case of a negative. It is possible to input a negative evaluation by natural operation. Thereby, the user can input evaluation by intuitive operation.

  When a negative evaluation is input, the control unit 11 proceeds to the process of step S3 to notify the next phoneme text in order to prompt the evaluation of the next recognition candidate phoneme text. In this case, the processes of steps S3 to S5 are repeated until a positive evaluation is input, and the phoneme texts of the recognition candidates are notified in order. That is, from the viewpoint of the user, only by performing a negative evaluation input operation, the phoneme text of the next recognition candidate is automatically selected and notified, and there is no need to separately perform an operation of selecting the phoneme text.

  On the other hand, when an affirmative evaluation is input, the control unit 11 performs processing corresponding to the phoneme text of the affirmative evaluation, for example, in the case of a car navigation device, the phoneme text is “Chizukakuai (map expansion)”. Executes a process of enlarging the displayed map, and in the case of a car audio device, if the phoneme text is “Shi Disai (CD playback)”, the CD playback process is performed. Execute. That is, the operation that the speech recognition apparatus 10 requests the user at the time of speech recognition is only a negative evaluation or a positive result input operation.

  Next, the control unit 11 performs statistical processing (correlation calculation between the combination of the recognition algorithm recognition parameters used and the frequency of “positive evaluation” and “negative evaluation” of the evaluation result) based on the evaluation result (step) S6) A combination of a recognition algorithm and a recognition parameter having a high correlation and a high frequency of “positive evaluation” is selected (step S7).

  And the control part 11 performs the speech recognition process (process of step S2) according to the combination of the recognition algorithm selected in step S7, and the recognition parameter about the audio | voice input via the microphone 12, and a power supply is turned off. For example, the learning data stored in the condition storage memory 15 is updated (learning process) so that the combination of the recognition algorithm and the recognition parameter at that time is the default (step S8). Thereby, a voice recognition algorithm and a recognition parameter can be improved in real time reflecting a user's evaluation result. The above is the speech recognition procedure of the speech recognition apparatus 10.

  As described above, when a negative evaluation is input, the speech recognition apparatus 10 according to the present embodiment notifies the next recognition candidate phoneme text and prompts the user to input the evaluation of the text. There is no need to perform an operation (selection operation or determination operation) for selecting a phoneme text from recognition candidates, and only an evaluation input operation is required. Furthermore, the voice recognition device 10 is easy and natural because the input operation for affirmative evaluation is an operation for the user to stroke the touch pad 21 and the input operation for a negative result is an operation for the user to tap the touch pad 21. Evaluation can be input by operation. Thereby, fatigue and stress due to input are reduced. Thus, if the user knows the position of only the touch pad 21 to be operated, the user can easily operate without relying on vision.

  In addition, embodiment mentioned above shows the one aspect | mode of this invention, This invention is not limited, It can change arbitrarily within the scope of the present invention. Further, the configuration shown in the embodiment is not limited to the configuration, and can be appropriately changed without departing from the gist of the present invention. For example, in the above-described embodiment, the case where the positive evaluation and the negative evaluation are input by the operation of stroking the touch pad 21 and the operation of hitting the touch pad 21 has been described. What is necessary is just to be able to input negative evaluation, and other operation methods may be applied.

  In the above-described embodiment, when an evaluation is input, sound may be emitted from the speaker 17 according to the evaluation. Specifically, for example, when a negative evaluation is input, an error word is synthesized by speech synthesis, and when a positive evaluation is input, a joy word is synthesized by voice synthesis and the response to the evaluation May be performed. Thereby, the user can perform an operation in an interactive manner, and the stress due to the incompleteness of the speech recognition process is alleviated.

  In the above-described embodiment, the case where the present invention is applied to the in-vehicle voice recognition device 10 has been described. However, the in-vehicle voice recognition device (for example, a personal computer in which a voice recognition application program is installed). It is applicable to.

It is a block diagram which shows the outline | summary of the speech recognition apparatus concerning this embodiment. It is a top view of the user evaluation input device of a voice recognition device. It is a sectional side view of a user evaluation input device. It is a figure where it uses for description of input operation of affirmation evaluation. It is a figure where it uses for description of input operation of negative evaluation. It is a flowchart which shows the procedure of voice recognition.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Speech recognition apparatus 11 Control part 12 Microphone 16 Speech synthesis part 17 Speaker 20 User evaluation input device 30A-30D Electrode

Claims (10)

Speech recognition means for recognizing a plurality of candidate phoneme texts from a speech signal;
Informing means for informing one phoneme text of a plurality of candidates recognized by the speech recognition means,
An input device for inputting a positive or negative evaluation of the phoneme text notified by the notification means,
When a negative evaluation is input through the input device, the notification unit notifies other phoneme texts of the plurality of candidate phoneme texts in order to prompt input of evaluation of other phoneme texts. Voice recognition device.   The speech recognition apparatus according to claim 1, wherein the input device can input the positive evaluation and the negative evaluation by different operation modes. Speech recognition means for recognizing phoneme text from speech signals;
Informing means for informing the phoneme text recognized by the speech recognition means,
An input device for inputting a positive or negative evaluation of the phoneme text notified by the notification means,
The speech recognition apparatus, wherein the input device can input the positive evaluation and the negative evaluation by different operation modes.   The speech recognition apparatus according to claim 2, wherein the operation mode of the positive evaluation is a stroke operation.   The voice recognition device according to claim 4, wherein the input device includes an operation body provided so as to be swingable, and detects the stroke operation by detecting the swing of the operation body.   The voice recognition apparatus according to claim 2, wherein the negative evaluation operation mode is a tapping operation.   The voice recognition device according to claim 6, wherein the input device includes an operation body provided so as to freely move up and down, and detects the tapping operation by detecting up and down movement of the operation body.   The speech recognition apparatus according to claim 1, further comprising a sound emitting unit that emits a sound in accordance with an evaluation input via the input device.   The speech recognition apparatus according to claim 1, further comprising a process execution unit that executes a process according to the phoneme text when the positive evaluation is input. Comprising storage means for storing a plurality of recognition algorithms and a plurality of recognition parameters;
The voice recognition means performs voice recognition based on any of the recognition algorithms and recognition parameters stored in the storage means, and the combination of the recognition algorithm and the recognition parameters used for the voice recognition, Based on the frequency of positive evaluation and negative evaluation input in the above, a combination of a recognition algorithm and a recognition parameter having a high correlation and a high positive evaluation is determined, and thereafter, speech recognition is performed with this determined combination. The speech recognition apparatus according to claim 1.

Images