JP2003099089A - Speech recognition/synthesis device and method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a speech recognition/synthesis device in which it is unnecessary to previously register vocabularies being key words and precision in speech recognition is high. SOLUTION: The device is provided with a text analysis section 1 which is used to analyze inputted character strings, a speech synthesis section 2 which generates synthesized speech corresponding to the character strings based on the analysis result of the section 1, a recognition vocabulary generating section 3 which generates vocabulary that is to be used for speech recognition based on the analysis result of the section 1, a vocabulary storage section 4 which stores the vocabulary generated by the section 3 and a speech recognition section 5 which recognizes inputted speech by referring to the vocabulary stored in the section 4. Basically, the device voice recognizes the language that is same as the language being read in synthesized speech.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition / synthesis apparatus and method for performing a voice recognition process for recognizing an input voice and a voice synthesis process for converting a character string of a text into a voice, and such a method. Program for realizing various devices and methods.

[0002]

2. Description of the Related Art Due to the spread of information processing devices such as personal computers and the development of voice synthesis technology, the number of scenes in which a character string is converted into a voice and utilized is increasing. Since the information can be obtained from the ear without looking at the characters, it is effective when the display area is limited, and it is possible to know the information by voice while performing another work.

However, the usable range is limited only by the one-way processing of synthesizing and outputting the voice. If there is an interaction element such that the user works on the synthetic voice, the application is expected to expand further.
As a means of interacting with the user, there is a method of pressing a button or selecting from a display menu by key input, but it is natural for the user to respond to the synthesized voice with a voice, which is convenient for the user. It is highly likely.

Many techniques for allowing a user to answer a voice inquiry by voice have been developed as voice conversation techniques. However, the technology of reading an arbitrary sentence by text-to-speech synthesis and recognizing the user's voice uttered in response to it is different from the voice dialogue.

As a voice recognition technique for text read aloud by voice synthesis, there is one disclosed in Japanese Patent Laid-Open No. 6-338946. According to the disclosed technology, voice reproduction can be performed from a position desired by a user according to information about a keyword uttered by the user during a voice output period of a mail.

[0006] For example, "... contacting the meeting. Well, the date and time is from 9:00 on April 1. The place is ..."
When the synthetic voice is output, the utterance "date and time" can be recognized and the synthetic voice after "date and time" can be output again.

In order to implement the technique of Japanese Patent Laid-Open No. 6-338946, the components shown in FIG. 12 are required. The text analysis unit 51 performs a morphological analysis of the input character string to obtain reading information and the like. The speech synthesis unit 52 obtains prosody information based on the result of the text analysis processing and generates synthetic speech. The vocabulary storage unit 54 stores a keyword to be re-outputted as a voice, as a voice recognition dictionary. The voice recognition unit 55 refers to the dictionary stored in the vocabulary storage unit 54 to perform voice recognition.

Only the minimum vocabulary necessary for the voice re-output operation may be registered in the vocabulary storage unit 54, or a large number of unspecified large vocabularies may be registered. In any case, a predetermined fixed vocabulary is registered in the dictionary, and the voice recognition unit 55 operates to obtain a recognition result by performing a matching process between the input voice and the registered vocabulary. .

[0009]

However, the technique disclosed in the above-mentioned Japanese Patent Laid-Open No. 6-338946 has a problem that the vocabulary as a keyword must be registered in advance. Further, although it is described that a voice recognition device with no restriction on vocabulary may be used, if there is no restriction on vocabulary, the recognition performance will be low and it will be impractical. Furthermore, since the voice recognition is used only for the purpose of re-outputting the synthetic voice, the voice recognition cannot be used for other purposes, for example, selecting one from a menu spoken by the synthetic voice. There is a problem.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to perform a voice recognition / synthesis apparatus and method capable of performing voice recognition with high accuracy and using voice recognition for a plurality of purposes without having to previously register a vocabulary as a keyword. Is to provide the program.

[0011]

SUMMARY OF THE INVENTION The present invention is basically intended to solve the above problems by recognizing only the words read aloud by synthetic speech.

Specifically, a speech recognition / synthesis apparatus according to one aspect of the present invention uses a text analysis unit for analyzing an input character string, and the character string based on the analysis result by the text analysis unit. Based on the speech synthesis unit that generates the corresponding synthetic speech and the analysis result by the text analysis unit,
It is input by referring to a recognition vocabulary generation unit that generates a vocabulary used for speech recognition, a vocabulary storage unit that stores the vocabulary generated by the recognition vocabulary generation unit, and a vocabulary stored in the vocabulary storage unit. And a voice recognition unit for recognizing the voice.

In this voice recognition / synthesis apparatus, a vocabulary used for voice recognition processing is generated from a character string input to the text analysis unit, and the voice recognition processing is performed on the input voice by referring to this vocabulary. ing. Therefore, unlike the above-mentioned conventional technique, it is not necessary to register the vocabulary as a keyword in advance. Further, since the vocabulary used for the voice recognition processing is generated from the character string input to the text analysis unit, the user can perform voice recognition with high accuracy when the user inputs the voice read out by the synthetic voice.

The speech recognition / synthesis apparatus includes vocabulary expansion means for expanding the vocabulary stored in the vocabulary storage unit based on the similarity between words, and words other than the input character string are also included in the vocabulary storage unit. It may be stored in.

Further, the speech recognition / synthesis apparatus may further include a language likelihood estimation unit for obtaining a language likelihood.
In this case, the voice recognition unit performs voice recognition in consideration of the language likelihood. The language likelihood can be obtained based on the time-series order of the generated vocabulary, or can be obtained based on the importance (probability of utterance) of each word. In any case, the speech recognition performance is improved by considering the language likelihood at the time of speech recognition.

In one embodiment, the voice recognition / synthesis device further includes a menu selection unit. Then, one or more menu items as options are input to the text analysis unit, and the menu selection unit selects one from the one or more menu items based on the recognition result by the voice recognition unit. It is like this. That is, here, voice recognition is used for the purpose of menu selection.

In this case, the voice recognition / synthesis apparatus may further include a correspondence storage unit for storing the menu item and one or more words constituting the menu item in association with each other.
The menu selection unit can execute menu selection if the voice recognition result by the voice recognition unit and at least a part of the menu item input to the text analysis unit at least partially match.

In another embodiment, the voice recognition / synthesis device further includes a link information acquisition unit. Then, hypertext is input to the text analysis unit, and the link information acquisition unit obtains information on the link destination of the hypertext based on the voice recognition result by the voice recognition unit. Here, speech recognition is used for the purpose of linking.

In this case, the voice recognition / synthesis apparatus may further include a correspondence storage unit for storing the link character string name and one or more words forming the link character string name in association with each other. The link information acquisition unit executes acquisition of information of a link destination when the voice recognition result by the voice recognition unit and at least a part of the link character string name input to the text analysis unit at least partially match. be able to.

The voice synthesizing section may include means for changing the voice synthesizing mode depending on the presence or absence of a link. Since the link information and the other text information are output as different synthetic voices, the user can determine which character string in the read text is linked by the difference in the synthetic voices.

The voice recognition / synthesis apparatus of the present invention having any one of the above configurations can be applied to a mobile terminal device.

Further, the speech recognition / synthesis apparatus of the present invention having any one of the above configurations can be applied to a communication system having a server and a terminal. In this case, according to another aspect of the present invention, of the constituent parts of the speech recognition / synthesis apparatus, the text analysis unit and the recognized vocabulary generation unit are provided on the server side, and the other constituent parts are the terminals. It is provided on the side.

The present invention further includes a step of analyzing the input character string, a step of generating a synthetic voice corresponding to the character string based on the result of the analysis, and a voice recognition based on the result of the analysis. Speech recognition / synthesis characterized by including a step of generating a vocabulary to be used, a step of storing the generated vocabulary, and a step of recognizing an input voice with reference to the stored vocabulary. Provide a way.

The present invention further provides a program for causing a computer to execute each step of the above speech recognition / synthesis method, and a computer-readable recording medium recording the program.

[0025]

BEST MODE FOR CARRYING OUT THE INVENTION The present invention will be described in detail below with reference to the embodiments shown in the drawings.

<First Embodiment> FIG. 1 is a block diagram showing the arrangement of a speech recognition / synthesis apparatus according to the first embodiment. The text analysis unit 1 performs a morphological analysis on the input character string to obtain reading information and the like. Speech synthesizer 2
Generates prosody information based on the result of text analysis processing and generates synthetic speech. The recognition vocabulary generation unit 3 generates a recognition vocabulary that can be used in voice recognition based on the result of the text analysis process, and creates a dictionary including this recognition vocabulary. The vocabulary storage unit 4 stores the dictionary created by the recognized vocabulary generation unit,
It is referred to by the voice recognition unit 5 during voice recognition. The voice recognition unit 5 refers to the dictionary recorded in the vocabulary storage unit 4 to perform voice recognition. This voice recognition / synthesis apparatus is configured by a computer equipped with a program for performing the processing shown in FIG.

When the configuration of the first embodiment shown in FIG. 1 is compared with the configuration of the conventional speech recognition / synthesis apparatus shown in FIG. 12, the speech recognition vocabulary is obtained from the text analysis result in the first embodiment. It can be seen that the recognition vocabulary generation unit 3 for generating is newly added. Either a predetermined keyword is registered in the vocabulary storage unit 54 in FIG. 12 or a large vocabulary without limitation is registered. In the first embodiment, the recognition vocabulary generation unit 3 that generates the recognition vocabulary that can be used in the voice recognition based on the result of the text analysis processing is provided, so that it is in a one-to-one correspondence with the words of the text read aloud. Since a corresponding dictionary can be created, it is possible to perform voice recognition with the minimum necessary recognition vocabulary, and it is possible to realize highly accurate voice recognition.

The flow of processing in the voice recognition / synthesis apparatus of FIG. 1 will be described with reference to the flowchart of FIG.

First, one sentence is acquired (S1), and the text analysis unit 1 performs text analysis (S2). The text analysis unit 1 analyzes the language of the input character string and divides it into the morphemes forming the character string. If there are multiple division candidates, all of them are output. Each candidate is given a likelihood that represents the likelihood of that candidate.

The text analysis technique required for speech synthesis is generally morphological analysis. Regarding morphological analysis processing,
The method using the longest right-matching method in the right direction or the connection table is generally used, and "Basics of Natural Language Analysis" (Hozumi Tanaka: Sangyo Tosho)
1989) and other documents are known.

Next, a reading is added to the divided morphemes. If there are multiple readings, output them all. Each reading is given a likelihood representing the degree of possibility of the reading.

For example, when a character string "weather of tomorrow" is given, it is divided into morphemes "tomorrow (noun)", "no (particle)" and "weather (noun)". There are two types of readings tomorrow, tomorrow and tomorrow, and the likelihood is given.

Next, recognition vocabulary generation is performed based on the result of the text analysis (S3). Here, information necessary for voice recognition is generated. The necessary information includes information about reading, information about likelihood of reading, information about adjacency probability of morphemes, and the like.

The information on the reading means that the phoneme symbol string "tomorrow" is "asu" and "ash ita", and the information on the likelihood of reading is "tomorrow".
Notation is "asu" or "a sh ita"
What is the probability of being, and the adjacency probability of morphemes is what the probability of "to" is next to "tomorrow".

Next, it is determined whether or not there is an input from the microphone (S4). The determination method may be based only on the level of the input signal, or when a talk switch is provided and pressed, the input for a few seconds after that may be determined as the voice recognition input. It is also possible to judge that the input while the talk switch is being pressed is the voice recognition input.

If it is determined that there is an input from the microphone, a voice recognition process is performed (S7). The method of speech recognition is described in detail in documents such as "Digital Speech Processing" (Furui: Tokai University Press, 1985).

The voice recognition is performed on the recognition vocabulary generated by the recognition vocabulary generating unit 3, and the recognition result is presented (S8). The method of presenting the recognition result may be a method of displaying the result on the screen or a method of outputting the synthetic voice. In the case of outputting the recognition result by the synthetic voice, it is possible to take a method of temporarily stopping the text being read aloud and restarting the presentation of the recognition result as soon as it is finished.

The analysis result obtained by the text analysis unit 1 is
It is used not only for creating a voice recognition dictionary but also for voice synthesis. In the case of speech synthesis, since the reading must be uniquely determined, only one candidate having a high likelihood is selected from words having a plurality of readings, such as “tomorrow”.
This point is different from the process of recognition vocabulary generation in which a plurality of readings are all registered as a vocabulary.

The text analysis unit 1 sends information about the most probable reading to the speech synthesis unit 2 based on the obtained word and the likelihood of reading. The voice synthesizing unit 2 obtains prosody information and generates and outputs synthetic voice (S5). Regarding the technology of speech synthesis, literatures such as "Digital Speech Processing" (Furui: Tokai University Press, 1985) are known.

As a result of the discrimination processing in S6, if there is a next sentence,
The process returns to S1 and the same processing is repeated. If there is no next sentence, the process ends.

By providing the recognized vocabulary generation unit 3 shown in FIG. 1 with a vocabulary expansion function, vocabularies other than the character string read aloud by text-to-speech synthesis can be registered as recognized vocabularies. The vocabulary extension function will be described below.

A dictionary describing the similarity of a plurality of words is generated in advance and stored in a storage unit (not shown). The lexical similarity indicates how similar the words A and B are to each other semantically and pragmatically.

The classification vocabulary table (National Institute for Japanese Language) shows synonymous words classified by hierarchy and can be considered as an example of a dictionary describing similarities. That is, according to the classification vocabulary table, another word having the same meaning as the word A can be obtained.

Using this information, when the word A appears in the read-aloud text, the recognition dictionary generator 3 can also generate another word B having the same meaning as A as the voice recognition vocabulary, and the vocabulary expansion can be realized.

Besides determining whether the words A and B are synonymous,
By determining whether or not the layers to which the words A and B belong are synonymous, it is possible to expand more vocabulary.

The classification vocabulary table is data describing whether two words have the same meaning or not, and it can be considered that the degree of the same meaning is binarized (discretized).
By digitizing the degree of similarity continuously, more flexible vocabulary expansion can be realized.

<Second Embodiment> FIG. 2 is a block diagram showing the arrangement of a speech recognition / synthesis apparatus according to the second embodiment. The text analysis unit 11, the voice synthesis unit 12, the recognized vocabulary generation unit 13, the vocabulary storage unit 14, and the voice recognition unit 15 have the same functions as the constituent units 1, 2, 3, 4, and 5 shown in FIG. 1, respectively. Have. The speech recognition / synthesis apparatus according to the second embodiment differs from the speech recognition / synthesis apparatus shown in FIG. 1 in that the menu selection unit 16 is provided.

The menu selection unit 16 selects a menu based on the recognition result by the voice recognition unit 15.

For example, "Today's news, tomorrow's weather,
"Today's TV ..." is a menu, and you can get news information by selecting "Today's News".
It is assumed that there is a device (computer) that can obtain weather information by selecting “Tomorrow's weather”. By incorporating the speech recognition / synthesis device of the present embodiment into such a device,
You can read out the menu and select menu items by voice. Specifically, the voice synthesizer 12 reads aloud the above-mentioned menu "Today's news, tomorrow's weather, today's TV, ...", and the user who is listening to it says "Tomorrow's weather". Then, the voice recognition unit 5 recognizes the vocabulary, and the menu selection unit 16 selects the menu item "weather of tomorrow" based on the recognition result. As a result, information such as "Tomorrow's weather is sunny" can be presented.

As a method of presenting information, characters may be displayed on the screen, or synthetic speech may be generated and output.

<Third Embodiment> FIG. 3 is a block diagram showing the arrangement of a speech recognition / synthesis apparatus according to the third embodiment. The text analysis unit 21, the voice synthesis unit 22, the recognized vocabulary generation unit 23, the vocabulary storage unit 24, the voice recognition unit 25, the menu selection unit 26, and the corresponding constituent units 11, 12, 1 in FIG.
It has the same function as 3, 14, 15, 16. The third embodiment further includes a correspondence storage unit 27, and the selection of menu items by the menu selection unit 26 is
The second point is that it is performed based on the data from the correspondence storage unit 27.
Different from the embodiment.

The correspondence storage unit 27 stores the correspondence relationship between the menu item and each word constituting the menu item.

By referring to the correspondence storage unit 27, even if the user utters only a part of the words constituting the menu item, the menu selection unit 26 can select the menu item. For example, as in the case of the second embodiment 2, "Today's news, tomorrow's weather, today's TV" is a menu, and selecting the item "Today's news" will give news information, It is assumed that there is a device that can obtain weather information by selecting the item “weather of tomorrow”. When the voice recognition / synthesis device according to the present embodiment is incorporated in this device, the correspondence storage unit 27 configures menu items such as “Today's menu”, “Tomorrow's weather”, “Today's TV”, and each menu. The word "today"
The correspondence relationship between "no", "menu", "tomorrow", "no", "weather", "today", "no", and "television" is stored. An example of the contents of the correspondence storage unit 27 in this case is shown in FIG.

When the user utters "news", the voice recognition unit 25 recognizes the vocabulary "news" and checks the constituent words recorded in the correspondence storage unit 27. Since the menu item corresponding to “news” is “today's news”, this menu item is output from the correspondence storage unit 27 and selected by the menu selection unit 26.

When the menu is read out by voice, the user usually does not care what the unit of the menu (menu item) is, that is, "today's news" or "news". It is also difficult to utter a menu item that is too long.
Therefore, it is useful that the menu item can be recognized even if only some of the words that constitute the menu item as described above are uttered.

However, there are cases where one word is included in a plurality of menu items. For example, the word "today" appears in both "today's news" and "today's television." In this case, tell the user that there are multiple options,
This can be dealt with by providing a mechanism for selecting again.

<Fourth Embodiment> FIG. 4 is a block diagram showing the arrangement of a speech recognition / synthesis apparatus according to the fourth embodiment.
The text analysis unit 31, the voice synthesis unit 32, the recognized vocabulary generation unit 33, the vocabulary storage unit 34, and the voice recognition unit 35 have the same functions as the corresponding constituent units 1, 2, 3, 4, and 5 shown in FIG. . The fourth embodiment differs from the first embodiment in that a language likelihood estimation unit 38 is provided.

The language likelihood estimation unit 38 obtains the language likelihood based on the time-series order of the generated vocabulary. The obtained language likelihood is written in the vocabulary storage unit 34, and the speech recognition unit 35. Referenced by.

First, the time-series order of vocabulary will be described. The time-series order is the order of appearance of words in a character string, and is an index indicating how far the previously read word is from the point of time when the character string is read.

For example, it is assumed that the character string "Today's news, tomorrow's weather, today's TV ..." is input and "News" is currently read out. As shown in FIG. 8, the time sequence order is assigned such that "today" is the first word to be read aloud, "news" is the third word, and so on. In this example, the larger the number in the chronological order, the more recently read the word. When using words that have been speech-synthesized as a menu option, the larger the number in time-series order, the higher the probability of being uttered. This is based on the reason that words read aloud in the near past are preferentially spoken because they are more memorable than words spoken in the distant past. In particular, if you think of how to listen to synthetic speech while doing some other work and get the information by uttering the word like parrot when you can hear the content you are interested in, this tendency Becomes even stronger.

Therefore, it is necessary to provide a mechanism for giving priority to words read in the near past over words read in the distant past. To achieve this, language likelihood is used. The linguistic likelihood is adjusted according to the difference in the time-series order between the currently read word and the already read word. That is, the smaller the time-series order difference, the higher the language likelihood, and the larger the order difference, the lower the language likelihood.

A calculation example of the language likelihood is shown below. Generally, the language likelihood can be obtained from the appearance probability of the word.
The appearance probability P (w) is as follows, where N is the total number of words in the learning data and C (w) is the number of times the word w appears in the learning data. P (w) = C (w) / N Here, using the information of the time series order described above,
The above appearance probability is corrected. The correction value of the word w at a certain time t is α (w, t). α is calculated based on the time series order of the word w. The time-series order is the order of words read aloud in a text-to-speech voice, and the smaller the number, the more distantly the word is spoken in the past. Therefore, the smaller the time series order corresponding to w, the smaller α (w, t) is set.

FIG. 8 shows an example of the time-series order of words at a certain time t0. The state of FIG. 8 shows the time-series order at time t0 immediately after the reading of the word "news" in the character string "Today's news, tomorrow's weather, today's TV ..." is finished. The correction value α at this time has the following relationship. α (Today, t0) <α (,, t0) <α (News, t0) By taking in this α, the appearance probability of a certain word at a certain time point t when the synthetic speech is read is defined as follows. Note that the following definition shows an example of obtaining the appearance probability, and the present invention is not limited to this. P (w, t) = λ {(C (w) + α (w, t)) / N} where λ is a constant for making the total of the probability values one.

The above appearance probability has a different value at time t1 which is different from t0. For example, the state of FIG. 9 is a time-series state at time t1 immediately after the reading of the word "weather" is finished. The word "news" has a long past compared to t0, so the following relationship holds. α (News, t0) ＞ α (News, t1) As described above, by obtaining the appearance probability considering the time-series state and using this as the language likelihood, it is possible to use words with higher probabilities of utterance. Since the appearance probability can be increased, a highly accurate voice recognition function can be realized.

As a method of obtaining the language likelihood in the language likelihood estimation unit 38, the method considering the time series state has been described above, but there are other methods. For example, it is a method of preliminarily assigning a value indicating the ranking of importance of each word to a dictionary used for text analysis, and obtaining the language likelihood based on the value. Assigning a high rank value to a word that is likely to be uttered, that is, a word that is highly important, and reflecting that value in the language likelihood to give the recognition vocabulary a language likelihood according to the importance. You can As a result, it becomes possible to make recognition easier for words having higher importance, and it is possible to further improve recognition performance.

<Fifth Embodiment> FIG. 5A is a block diagram showing the arrangement of a speech recognition / synthesis apparatus according to the fifth embodiment. The text analysis unit 41, the voice synthesis unit 42, the recognized vocabulary generation unit 43, the vocabulary storage unit 44, the voice recognition unit 45, and the correspondence storage unit 47 are the corresponding constituent units 21, 22, 2 shown in FIG.
It has the same function as 3, 24, 25, 27. In the fifth embodiment, the link information acquisition unit 4 is used instead of the menu selection unit.
The point that 6 is provided is basically different from the third embodiment.

In the fifth embodiment, a document described in hypertext is acquired by using WWW (World Wide Web) or the like, a character string is read aloud by a synthetic voice, and a user who hears it reads a link. The character string of the anchor part (hereinafter, "link character string")
By uttering a name, it is possible to obtain information on the link destination. Therefore, this voice recognition / synthesis apparatus includes the link information acquisition unit 46, and stores the correspondence relationship between the link character string and each character string (word) forming the link character string in the correspondence storage unit 47. .

The text "Today's news, tomorrow's weather, today's TV ..." is hypertext acquired on the WWW, and today's news articles are written at the "Today's news" link. When the user says "Today's news" in the situation that the user is present, the content of the news can be acquired.

Below, HTML (HyperText Markup Langu
age hypertext markup language)
An example of a page description language represented by is described below. The link information refers to the description as shown in FIG. The link information includes a link destination 101 and a link character string 102. In the above-mentioned example, the link character string corresponds to a file in which the text "Today's news, tomorrow's weather, today's TV ..." is described, and the link destination is the content of "Today's news". URL (Uniform Resource Lo
Address information represented by cator uniform resource locator) is described. The file at this address contains the information you want to get. Figure 1
In 0, 103 and 104 are link destination files. The link character string is a character string that is normally displayed on the WWW browser, and here is a character string that is voice-synthesized by the voice synthesizer 42.

Normally, an HTML document is displayed on a WWW browser, and the HTML document described in the link destination can be displayed by clicking the link character string. However, in the present embodiment, voice recognition is performed regardless of the WWW browser. And necessary information can be acquired by using voice synthesis. This means that hypertext links can be followed only by auditory information, without using visual information. As a result, it becomes possible to obtain necessary information on the WWW by listening to the sound and giving an instruction by voice while performing other work on the personal computer or the like.

The mechanism for acquiring the information of the specified URL is the same as the method of acquiring by the WWW browser and is not directly related to the present invention, and therefore will not be described here.

By the way, in order to allow the user to determine which character string in the read text is linked, a voice for changing the voice synthesis style as shown in FIG. 5B. It is preferable to provide the synthesizing mode changing means 42a in the voice synthesizing unit 42.

In the case of visual information, it is generally possible for the user to determine whether the character string is linked or not by changing the character string color or font. is there. In the case of voice synthesis, the information can be transmitted to the user by changing the voice synthesis mode.

The voice synthesizing mode includes speakers of various voice qualities, speech rates, pitches, and powers. For example, in the case of a speaker, the presence or absence of the link destination can be notified by speaking the linked character string with a male voice and the other with a female voice.

Alternatively, the linked character strings are distinguished from each other by a method such as uttering at a slower speed than other character strings, uttering at a higher pitch, or uttering with increased power. You can also

As described above, by changing the voice synthesis mode, it becomes possible to inform the user of the presence or absence of a link.

Although the correspondence storage unit 47 is provided in FIGS. 5A and 5B, it may be omitted.

<Sixth Embodiment> Any of the above-described embodiments and a modification of the voice recognition / synthesis apparatus can be mounted on a terminal device such as a portable terminal device. At that time, it is also possible to have a configuration in which the text analysis unit and the vocabulary generation unit are provided on the server side and the other parts are provided on the terminal side. Figure 11
As an example of such a configuration, a configuration of a communication system corresponding to a configuration in which all the components of the first embodiment are divided into a server side and a terminal side is shown. In the example of FIG. 11, the text analysis unit 61 out of all the components of the speech recognition / synthesis apparatus.
The recognition vocabulary generation unit 63 is provided on the server side, and the terminal side includes only the voice synthesis unit 62, the vocabulary storage unit 64, and the voice recognition unit 65.

Although the present invention has been described with reference to various embodiments, the configurations of the above-described embodiments may be combined or modified as appropriate. For example,
The language likelihood estimation unit shown in FIG. 4 may be provided in the speech recognition / synthesis apparatus shown in FIGS. 1, 2, 3 or 5 in order to improve the accuracy of speech recognition.

[0080]

As is clear from the above, the speech recognition / synthesis apparatus according to the present invention does not need to have a predetermined keyword or a wide range of words as a dictionary, and can use only a character string read aloud by synthetic speech. Since the voice recognition dictionary is created automatically, high performance voice recognition can be realized.

Further, by obtaining the language likelihood in consideration of the time series state, it is possible to realize the speech recognition with higher performance.

Further, by being able to recognize the same words as the contents read aloud by the synthetic voice, it is possible to execute the menu selection and the like without visual inspection.

Also, while doing other work on the personal computer, the synthesized voice is played, and when a word of concern is heard,
If you say that parrot back and say it, you will be able to obtain the desired information, and you will be able to use it in a "while" way, such as obtaining the necessary information while doing other work.

By implementing the present invention, it becomes possible to use an apparatus that has conventionally been used only for one purpose, for a plurality of purposes at the same time, and to provide an environment in which a plurality of works can be efficiently performed in a fixed time. Become.

[Brief description of drawings]

FIG. 1 is a first embodiment of the present invention;
It is a block diagram of a synthesizer.

FIG. 2 is a second embodiment of the present invention;
It is a block diagram of a synthesizer.

FIG. 3 is a third embodiment of the present invention; speech recognition /
It is a block diagram of a synthesizer.

FIG. 4 is a fourth embodiment of the present invention;
It is a block diagram of a synthesizer.

5A is a block diagram of a voice recognition / synthesis device according to a fifth embodiment of the present invention, and FIG. 5B is a block diagram showing a modified example.

FIG. 6 is a flowchart showing an example of the flow of processing of the present invention.

FIG. 7 shows an example of a correspondence storage unit.

FIG. 8 shows an example of a relationship between words and a time series order.

FIG. 9 shows an example of a relationship between words and a time-series order.

FIG. 10 shows an example of link information.

FIG. 11 is a block diagram of a communication system to which the speech recognition / synthesis device of the present invention is applied.

FIG. 12 is a block diagram of a conventional voice recognition / synthesis device.

[Explanation of symbols]

1, 11, 21, 31, 41, 51, 61 ... Text analysis part, 2, 12, 22, 32, 42, 52, 62 ... Speech synthesis part, 3, 13, 23, 33, 43, 63 ... Recognition vocabulary Generation unit, 4, 14, 24, 34, 44, 54, 64 ... Vocabulary storage unit, 5, 15, 25, 35, 45, 55, 65 ... Speech recognition unit 16, 26, 36 ... Menu selection unit 27, 37 , 47 ... Corresponding storage unit 38 ... Language likelihood estimation unit 46 ... Link information acquisition unit 101 ... Link destination 102 ... Link character strings 103, 104 ... Link destination file

─────────────────────────────────────────────────── ─── Continued Front Page (51) Int.Cl. ⁷ Identification Code FI Theme Coat (Reference) G10L 15/18 G10L 3/00 E 15/28 551P

Claims (13)

[Claims] 1. A text analysis unit that analyzes an input character string, a voice synthesis unit that generates a synthetic voice corresponding to the character string based on an analysis result by the text analysis unit, and a text analysis unit. Based on the analysis result, a recognition vocabulary generation unit that generates a vocabulary used for speech recognition, a vocabulary storage unit that stores the vocabulary generated by the recognition vocabulary generation unit, and a vocabulary stored in the vocabulary storage unit. A voice recognition / synthesis device comprising: a voice recognition unit for recognizing an input voice with reference. 2. The voice recognition / synthesis apparatus according to claim 1, further comprising a vocabulary expansion unit for expanding the vocabulary stored in the vocabulary storage unit based on the similarity between words, and other than the input character string. A voice recognition / synthesis apparatus characterized in that words are also stored in the vocabulary storage unit. 3. The speech recognition / synthesis apparatus according to claim 1, further comprising a language likelihood estimation unit that obtains a language likelihood, wherein the speech recognition unit performs speech recognition in consideration of the language likelihood. A voice recognition / synthesis device characterized by performing. 4. The voice recognition / synthesis apparatus according to claim 1, further comprising a menu selection unit, wherein the text analysis unit is input with one or more menu items as options. The voice recognition / synthesis apparatus, wherein the menu selection unit selects one from the one or more menu items based on a recognition result by the voice recognition unit. 5. The voice recognition / synthesis apparatus according to claim 4, further comprising: a correspondence storage unit that stores a menu item and one or more words constituting the menu item in association with each other, wherein the menu selection unit includes: A voice recognition / synthesis apparatus, which performs menu selection if a voice recognition result by the voice recognition unit and at least a part of a menu item input to the text analysis unit match. 6. The voice recognition / synthesis apparatus according to claim 1, further comprising a link information acquisition unit, wherein hypertext is input to the text analysis unit, and the link information acquisition unit. A voice recognition / synthesis apparatus, characterized in that, based on a voice recognition result by the voice recognition unit, information on a link destination of the hypertext is obtained. 7. The speech recognition / synthesis apparatus according to claim 6, wherein the speech synthesis unit includes means for changing a speech synthesis pattern depending on the presence / absence of a link. 8. The voice recognition / synthesis apparatus according to claim 6 or 7, further comprising a correspondence storage unit that stores a link character string name and one or more words forming the link character string name in association with each other. The link information acquisition unit executes acquisition of information of a link destination when the voice recognition result by the voice recognition unit and at least a part of the link character string name input to the text analysis unit at least partially match. A speech recognition / synthesis device characterized in that 9. A communication system comprising the speech recognition / synthesis device according to claim 1, wherein the text analysis unit is one of the constituent parts of the speech recognition / synthesis device. A communication system, wherein the recognition vocabulary generation unit is provided on the server side, and the other components are provided on the terminal side. 10. A mobile terminal device comprising the voice recognition / synthesis device according to claim 1. Description: 11. A step of analyzing an input character string, a step of generating a synthetic voice corresponding to the character string based on a result of the analysis, and a voice recognition based on a result of the analysis. A voice recognition / synthesis method comprising: a step of generating a vocabulary; a step of storing the generated vocabulary; and a step of recognizing an input voice with reference to the stored vocabulary. 12. A step of analyzing a character string input to a computer, a step of generating a synthetic voice corresponding to the character string based on a result of the analysis, and a voice recognition based on a result of the analysis. A program for executing a step of generating a vocabulary to be used, a step of storing the generated vocabulary, and a step of recognizing an input voice with reference to the stored vocabulary. 13. A computer-readable recording medium in which the program according to claim 12 is recorded.