JPH09230889A - Speech recognition and response device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable a user to effectively respond and deal with the input of a word out of the registration uttered by the user erroneously without knowing it so that the inputted word is out of the registration by providing an external dictionary holding means in addition to an internal dictionary holding means to hold words corresponding to conventional registered words. SOLUTION: This device is provided with an internal dictionary holding means 104 to hold a registered vocabulary dependent on an application, and a external dictionary holding means 105 to hold a vocabulary out of the registration which are of the same kind as the registered vocabulary and are independent of the application, and the application control means 101 put together a vocabulary held by the internal dictionary holding means 104 and a vocabulary held by the external dictionary holding means 105 and transfers them to a speech recognition means 102 as a recognition object vocabulary. When the recognition result obtained from the speech recognition means 102 is included in the vocabulary held by the external dictionary holding means 105, the application control means 101 shows the result of the recognition to the user via a response output means 102.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information processing apparatus, and more particularly to a voice for improving a dialogue operation between a user and the apparatus when voice recognition is used as a data input means of a man-machine interface. It is related to recognition and utilization technology.

[0002]

2. Description of the Related Art A voice recognition response device is a device which is supposed to perform some task such as information retrieval. When a user inquires by voice, the result is shown by voice, and in some cases, by characters or images.

In a conventional voice response device, a target vocabulary that can be recognized by voice in a certain scene is one vocabulary within a pre-registered range, and even if words of the same type (for example, a person's name) are registered in advance. Unrecognized words can never be recognized correctly. In this case, the voice response device shows a result of erroneously recognizing a similar word in the registered words, or a result of not corresponding (rejection) in the registered word. Even if the user utters repeatedly, the result is the same.

On the other hand, from the user's point of view, whether the word that is erroneously recognized or rejected is simply a word that cannot be recognized due to poor performance of the device, or whether the user's utterance is bad,
Or, there is no way to know if it is another factor, and even though I feel the inconvenience of the system, I repeat utterances many times and probably give up after a while and stop using.

[0005]

SUMMARY OF THE INVENTION In the conventional voice recognition responding apparatus as described above, there is a problem to deal with the case where a word not registered is input.

An object of the present invention is to provide an external dictionary holding means in addition to the conventional internal dictionary holding means for holding a word corresponding to a registered word, so that the unregistered word uttered by the user erroneously or unknowingly. It is an object of the present invention to provide a voice recognition responding device capable of responding to an input such as that the input word is a non-registered word by responding effectively to the user.

[0007]

According to the present invention, the vocabulary held by the internal dictionary holding means composed of the registered vocabulary depending on the application and the vocabulary of the same kind as the registered vocabulary but independent from the application are registered. The recognition target vocabulary is combined with the vocabulary held by the external dictionary holding unit composed of outside vocabulary, and the recognition target vocabulary is passed to the voice recognition unit, or the recognition result obtained from the voice recognition unit is included in the external dictionary. Provided is a voice recognition response device provided with application control means for transmitting the content of the word to the response output means in the case of a word to be displayed.

[0008]

FIG. 1 is a block diagram showing an embodiment of a voice recognition response device of the present invention. The application control means 101 combines the vocabulary held in the internal dictionary holding means 104 (hereinafter referred to as an internal dictionary) and the vocabulary held in the external dictionary holding means 105 (hereinafter referred to as an external dictionary) to be a recognition target vocabulary. To the voice recognition means 102. The internal dictionary and the external dictionary are paired, and a plurality of pairs may be provided. When the application can set a scene, a pair of an internal dictionary and an external dictionary is prepared for each scene, and a different internal dictionary and external dictionary are sent to the voice recognition unit 102 for each scene. Here, the application control means 10
For No. 1, the internal dictionary and the external dictionary are simply combined and sent to the voice recognition unit 102 as one dictionary (recognition target vocabulary 201). The contents of the dictionary may be, for example, hiragana notation of words as shown in FIG. The present invention is not limited to this notation and may be used as it is as an input of the voice recognition means 102 or can be converted into a form of a recognized vocabulary.

The voice 202 of the user is the voice recognition means 10.
Recognized by 2. Voice recognition means 102 in the present invention
Inputs the user's voice 202 and a list of words to be recognized (recognition target vocabulary 201), and recognizes one or more probable words in the list as the recognition result 2
Output as 03. When the recognition vocabulary is written in hiragana, the voice recognition process is performed using, for example, H with a syllable as a unit model.
It can be realized by MM (Hidden Markov Model).
When a syllable is used as a unit model, one syllable can be associated with one character in hiragana. However, the present invention does not limit how to determine the unit model. The voice recognition process by the HMM is known, and the details are omitted. For example, it is described in detail in "Nakakawa: Speech Recognition by Probability Model, IEICE".

The voice recognition means 102 recognizes the recognition target vocabulary 20.
It is not necessary to treat the internal dictionary and the external dictionary as 1, in particular. In other words, the recognition result 203 output from the voice recognition unit 102 does not need to be treated as if the internal dictionary and the external dictionary were particularly distinguished. The recognition result 203 is passed to the application control means 101. The application control means 101 checks whether the recognition result 203 is included in the internal dictionary or the external dictionary. Recognition result 2
If 03 is included in the internal dictionary, the process proceeds to the next process of the application. For example, as is possible with a conventional voice recognition response device, a response (for example, "Is it Saitou?") For confirming the recognized content to the user is performed via the response output means 103. On the other hand, when the recognition result 203 is not included in the internal dictionary but included in the external dictionary, the response output unit 103 makes a response indicating that the input is an unexpected wrong input other than registration. For example, when the application control unit 101 expects the name registered in the internal dictionary holding unit 104 in FIG. 2 for the input voice, the name “Suzuki” registered in the external dictionary holding unit 105 When the device is obtained as a recognition result, the device outputs a response to the user such as "Suzuki is not registered here" as the response output means 103.
Output via. By referring to this message, the user knows that the voice uttered by him / her is not defective, increases the reliability of the system, and can perform a smooth interactive operation.

As another case, when the word uttered by the user is not registered in either the internal dictionary or the external dictionary, the recognition result is shown by a program (not shown) included in the application control means 101. Register a new one in the external dictionary. The fact that the word uttered by the user is not registered in either the internal dictionary or the external dictionary means that consecutive misrecognitions have been repeated a certain number of times or more (Japanese Patent Laid-Open No. Sho 61-96).
62-39899), or when there is no recognition result repeated more than a certain number of times, or when the user selects the registered words in the internal dictionary and external dictionary in order to let the user select any of them. It can be determined depending on the cases such as not having.

FIG. 3 shows the procedure of the interactive operation between the user and the device when newly registering in the external dictionary, taking the name registration as an example.
Is shown in the flowchart of FIG. In FIG. 3, when the word uttered by the user is not registered in either the internal dictionary or the external dictionary, the response output means 103 of the apparatus says "There is no person with such a name". A message is output (step 301). Then, in order to newly register the unregistered name in the external dictionary, the device outputs a message to input the unrecognized name one by one. In order to check whether or not the user can handle, with an emphasis on interactive operations, for example, "Because a new name is registered in the dictionary, please enter the name you entered earlier one by one.
Do you have time Is output (step 302). If the user inputs an agreement such as "Yes" to this message (step 30)
3) Then, a new word is selectively acquired for each character from the user in response to monosyllabic recognition or presentation from the device (step 304), and newly registered in the external dictionary (step 305). If the user inputs a refusal such as "No" (step 303), a message such as "Please cooperate with new registration at the next opportunity if you like" is output to the user ( In step 306), the interactive operation at the time of new registration ends. The response output means 103
The input response content is converted into voice by voice synthesis processing and transmitted to the user, or displayed as characters and transmitted to the user. As described above, the voice recognition response device of the present invention can be realized.

FIG. 2 shows an example of the external dictionary and the internal dictionary according to the present invention. Specifically, it shows that when the internal dictionary is personal names, a list of personal names that frequently appear can be used as the external dictionary.

As another example of the case where the internal dictionary is a person's name, a list of readings in which the person's name included in the internal dictionary is easily misread can be used as the external dictionary.

Furthermore, as another example when the internal dictionary is personal names, a list of personal names included in the internal dictionary at any time in the past and deleted at a certain update time can be used as the external dictionary. For example, this is a situation that may occur due to a transfer.

Similar lists can be prepared for company names, department names, etc. in addition to personal names.

When the internal dictionary is a line name / station name of a railway company, a line name / station name list of another railway company can be used as an external dictionary.

With the above combinations, it becomes possible to flexibly deal with various voice inputs of unregistered words.

[0019]

According to the present invention, the voice recognition response device is
It becomes possible to deal with the voice input of a word not registered by the user, by recognizing the input and pointing out an error. In addition, the user becomes aware of the mistake and can re-word it correctly, or can re-word it with another expression, and the effect of improving the utilization efficiency of the device (such as shortening the time until completion of use) can be obtained.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a voice recognition response device according to the present invention.

FIG. 2 is a diagram showing an embodiment of an internal dictionary and an external dictionary according to the present invention.

FIG. 3 is a diagram showing an interactive operation procedure when registering an external dictionary according to the present invention.

[Explanation of symbols]

101 ... Application control means, 102 ... Voice recognition means, 103 ... Response output means, 104 ... Internal dictionary holding means, 105 ... External dictionary holding means, 106 ... Acoustic model holding means, 201 ... Recognition target vocabulary, 202 ... Speech, 203
... recognition result, 204 ... response content, 205 ... character, 206
…voice.

Claims (3)

[Claims] 1. A standard pattern of a recognition target vocabulary obtained by combining a given recognition target vocabulary and an acoustic model is compared with the acquired voice, and one or a plurality of recognition results are output. Speech recognition means, application control means for transmitting the recognition target vocabulary to the voice recognition means, and generating contents to be responded to the recognition result of the recognition target vocabulary, and response contents generated from the application control means In a voice recognition response device comprising a response output means for outputting as a voice or character information, an internal dictionary holding means for holding a registered vocabulary depending on an application, and a vocabulary of the same type as the registered vocabulary, which is registered independently of the application. External dictionary holding means for holding an external vocabulary, wherein the application control means is an internal dictionary The vocabulary held by the holding means and the vocabulary held by the external dictionary holding means are combined and passed to the voice recognition means as a recognition target vocabulary, and the application control means causes the recognition result obtained from the voice recognition means to be output to the outside. A voice recognition response device characterized in that when it is included in the vocabulary held by the dictionary holding means, the content to that effect is transmitted to the response output means. 2. A plurality of pairs of the internal dictionary holding means and the external dictionary holding means are provided, and the application control means selects and uses any one of the plurality of pairs according to the scene of the application. Item 1. A voice recognition response device according to item 1. 3. The vocabulary held by the external dictionary holding means is
The voice recognition response device according to claim 1 or 2, wherein the voice recognition response device can be updated even while being used.