JP3536524B2 - Voice recognition method and voice recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a speech recognition method and a speech recognition method.
More particularly, the present invention relates to a voice recognition method and a voice recognition device suitable for use in, for example, an electronic organizer for managing a schedule or an address book (name list).

[0002]

2. Description of the Related Art In a conventional device for storing and managing information such as an electronic organizer, an input device such as a keyboard, a pointing device (for example, a mouse, a pen, a touch panel, etc.) and an OCR (Optical Character Reader) is used. Information is input and stored using the device. Then, the stored information is searched in response to a request from the user, and the search result is output from an output device such as a display, a printer, or a speech synthesizer.

Such an electronic organizer can mainly manage a schedule, an address book, and the like. That is, for example, by inputting the date and time by operating the keyboard, or by pointing the date and time in the calendar displayed on the screen by the pointing device, the schedule registered and stored in advance and stored at that date and time is searched. It is displayed (or output as a synthesized sound). Furthermore, some electronic organizers automatically notify that date and time when a schedule is stored.

For example, when a last name or the like is input by operating a keyboard, an address or a telephone number of a person having the last name is searched from a pre-registered address book and displayed. If a plurality of pieces of information about a person with the same surname are stored and cannot be displayed on the screen, for example, information about such a person with the same surname is listed and displayed. , The user can scroll through the list,
Information of a desired person can be obtained.

When such an electronic organizer is configured using a keyboard having a large number of keys, a large display, or the like, an input by a keyboard, an input by instructing a display with a pointing device, and the like are not performed. Although it is easy to perform, it is difficult to make the device compact.

[0006] However, since the electronic organizer is usually used in a portable manner, it is required to have a small size. Therefore, there are methods of reducing the number of keys constituting the keyboard and reducing the size of the display, but this makes it difficult to input information and other information.

That is, for example, a keyboard is composed of four cursor keys for moving a cursor up, down, left, and right, and five keys of a decision key for instructing decision of a selection. In addition, the cursor is also displayed, the cursor is moved to a desired character position by operating a cursor key, and the input of the character is determined by operating a determination key. In some cases, the number of keys is very small compared to the number of characters that can be input, so it is necessary to operate the cursor key and the enter key to input one character, which makes the operation complicated. Become.
Further, an input of information, 1 perform each character is cumbersome, further, when it is not possible to display all the characters that can be input at one time on the display, when trying to enter the display that is not a character, the screen Must scroll so that the character is displayed.

[0008] Further, for example, there is a similar problem when a character displayed on a display is pointed by a pointing device.

The above problems occur not only when inputting information to be registered and managed, but also when searching for registered information. That is, for example, when searching the address book, it is possible to search not only the first and last names but also many search items such as postal codes, area codes, prefecture names in cities, and municipal names. When you do, many such search items are displayed on the display,
It is necessary to input a search item to be searched from among them with a small number of keys as described above.

[0010] Therefore, a device (electronic dictionary) having a built-in voice recognition device capable of inputting information by voice so that information and other data can be easily input is provided.
For example, JP-A-4-10166 and JP-A-4-40
557 and JP-A-4-42363.

According to such a device, the user can input various information by voice, and in this case, it is not necessary to perform complicated operations.

[0012]

By the way, as described above, when information is input by voice and the voice is recognized, all the information expected to be input are used.
For example, a dictionary (hereinafter referred to as a word)
If necessary, a dictionary in which such words to be subjected to speech recognition are registered must be registered in a recognition dictionary).

That is, for example, when inputting a schedule, it is expected that a meeting place or the like is input. Therefore, in order to be able to perform voice recognition, a place name indicating a place that may be input (for example, Tokyo, Kanagawa, Saitama, ..., Shinagawa, Shibuya, Yokohama City, Kawasaki City, etc.), station names (eg, Shinagawa Station, Shibuya Station, Yokohama Station, etc.), facility names (eg, Tokyo Tower, Tokyo Metropolitan Government, National Stadium, ...)
・ ・ Etc.) Everything else must be registered in the recognition dictionary.

However, if many speech recognition target words are registered in the recognition dictionary, the recognition performance is degraded, and it takes time from when a speech is input to when the speech recognition result is obtained. Instead, the user feels troublesome.

On the other hand, for example, in order to search for a schedule at a predetermined date and time, when the date and time are input, for example, the date and time based on the Christian era and the month or the year and year XX month
△ In addition to the designation by absolute date and time (hereinafter referred to as absolute date and time as appropriate) such as the date and time using the era such as the day, for example, based on the current date and time such as tomorrow, the day after tomorrow, next Monday, etc. In some cases, it is desired to specify the date and time (hereinafter, appropriately referred to as a relative date and time), but conventionally, it has been difficult to specify the relative date and time.

The present invention has been made in view of such a situation, and it is an object of the present invention to enable information and other inputs to be made easily and flexibly without increasing the size of the apparatus.

[0017]

According to a first speech recognition method of the present invention, an input step for inputting a character string consisting of one-letter words is performed.
And flop, a creation step of creating a continuous string by arranging a plurality of consecutive characters from string of character words, to consider a continuous string, characterized in that it comprises a speech recognizing speech recognition step .

The second speech recognition method according to the present invention uses a one-letter word
Input step to input a character string consisting of
In Ranaru string, a creation step of creating a continuous string by arranging a plurality of consecutive characters, a continuous string, by registering in the recognition dictionary, and a registration step of the speech recognition terms It is characterized by having.

The speech recognition apparatus according to the present invention provides a speech recognition target.
Record the recognition dictionary in which the speech recognition target words
Input the character string consisting of one-word words
Input means, and a continuous character string
Create a continuous character string by placing multiple characters
Register the continuation character string in the recognition dictionary as a speech recognition target word.
And a registration means.

In the first speech recognition method of the present invention,
Enter a character string consisting of one-letter words and send a sentence consisting of one-letter words
A continuous character string is created by arranging multiple consecutive characters from a character string.
Then, speech recognition is performed on the continuous character string.

In the second speech recognition method of the present invention,
Enter a character string consisting of one-letter words and send a sentence consisting of one-letter words
Consecutive character string by arranging multiple consecutive characters in the character string
Is created, and the continuous character string is registered in the recognition dictionary.
In this case, the word is used as a speech recognition target word.

In the speech recognition apparatus of the present invention, one character
Enter a character string consisting of words, and
Create a continuous character string by arranging multiple consecutive characters in
Then, the continuous character string is used as a recognition
To register.

[0023]

[0024]

[0025]

[0026]

[0027]

[0028]

[0029]

Embodiments of the present invention will be described below.
However, the components described in the claims and the embodiments of the invention
To illustrate the corresponding relationship with the specific example in
Become. This statement supports the invention described in the claims.
Specific examples to be described are described in the embodiments of the invention.
It is to confirm that. Therefore, the implementation of the invention
Although it is described in the form of
As there was a specific example not described here
However, that does not mean that the specific example
It does not mean that it will not respond. Reverse
Here, specific examples are listed here as corresponding to the configuration requirements.
Even if it had been
Does not correspond to configuration requirements other than configuration requirements
It does not mean.

Further, this description is based on the embodiments of the invention.
The invention corresponding to the described specific example is all described in the claims.
It does not mean that it is described. Paraphrase
For example, this description may include the components described in the embodiment of the invention.
It is an invention corresponding to a constitutional example and is described in the claims of this application.
Existence of invention not listed, that is, future divisional application
Denies the existence of inventions that have been added or amended
Not something.

In the voice recognition method according to claim 1,
Is a voice recognition method that recognizes the input voice.
Inputting a character string consisting of one-letter words
(For example, processing step S41 of the program in FIG. 11)
And multiple consecutive characters from a string of one-letter words
A creation step that creates a continuous string by placing (for example,
The processing step S46 of the program in FIG.
Speech recognition step (for example,
For example, processing step S42) of the program in FIG.
It is characterized by

In the voice recognition method according to the second aspect,
Recognizes the input speech as a speech recognition target.
Speech while referring to the recognition dictionary where the target words are registered
A speech recognition method that recognizes one-letter words.
An input step of inputting a column (for example, the program of FIG. 12)
Processing step S61) and a character string composed of one-letter words
A continuous character string is created by arranging multiple consecutive characters in
Creating step (for example, processing of the program of FIG. 12)
Management step S62) and register the continuous character string in the recognition dictionary.
Registration step as a speech recognition target word
(For example, processing step S69 of the program in FIG. 12)
And characterized in that:

[0033] In the voice recognition method according to claim 3,
Is a display step that displays a character string consisting of one-letter words
(For example, the display 3 in FIG. 3) and the speech
Displayed when a continuous character string registered in the
Of the continuous character string recognized by speech in the
Kana-Kanji conversion step for converting characters (for example,
Program processing step S49).
It is characterized by.

[0034] In the speech recognition apparatus according to claim 5,
Is a speech recognition device that recognizes input speech.
The speech recognition target words targeted for speech recognition are registered.
Storage means (for example, FIG. 3
Storage unit 13) and a character string composed of one-letter words
From input means (for example, microphone 2 in FIG. 3) and one-letter words
Arrange multiple consecutive characters in a character string
Create a character string and convert the continuous character string
Then, the registration means for registering in the recognition dictionary (for example, the system shown in FIG. 3)
And a stem control unit 12).

[0035]

FIG. 1 shows an external configuration of an embodiment of an electronic organizer to which the present invention is applied. This electronic organizer is portable so that it can be easily carried. A button 1 is provided on the right side thereof, and a microphone 2, a display 3, a speaker 4, and a cursor key are provided on a front panel. 5 are provided.

The button 1 is held by an elastic body such as a spring (not shown), and can be pressed by applying a force in the direction inside the apparatus. When the button 1 stops applying a force thereto, the button 1 returns to the original state by the elasticity of the spring. The button 1 is provided at a position where the electronic organizer can be easily operated with the thumb when the electronic organizer is held with the right hand.

The microphone 2 converts the sound inputted thereto into a sound signal as an electric signal. This voice signal is voice-recognized in the electronic organizer, and the voice recognition result is displayed on the display 3 or, if necessary, from the speaker 4.
It is designed to be output as a synthesized sound.

In the electronic organizer, button 1
While the button is pressed, the voice input to the microphone 2 is recognized. That is, in this embodiment, the voice section is not detected based on the output signal of the microphone 2 but the period during which the button 1 is pressed is set as the voice section. However, the voice section is detected by the electronic organizer based on, for example, the zero crossing or power of the output signal of the microphone 2, so that voice can be input without operating the button 1. It is possible to do so.

The display 3 is composed of, for example, an LCD (Liquid Crystal Display), a CRT (Cathod Ray Tube), or the like, and displays predetermined information. The speaker 4 is configured to output predetermined information as a synthesized sound, or to output a warning sound (for example, a so-called beep sound) for calling a user's attention.

The cursor keys 5 include an up key 5a and a down key 5
b, a left key 5c, and a right key 5d.
Up key 5a, down key 5b, left key 5c, or right key 5
d moves the cursor displayed on the display 3 up,
It is designed to be operated when moving down, left, or right, respectively. In this electronic organizer,
On display 3, along with some selectable items,
When the cursor is displayed, the cursor is moved to a position of a desired item by operating the cursor key 5, and the button 1 is pressed once and released (hereinafter, referred to as a button).
Clicking as appropriate) will confirm the selection of that item.

The display 3 can be integrally formed with, for example, a transparent tablet or the like.
In such a case, an item displayed on the display 3 can be selected by designating a display portion of the item with a fingertip, a pen, or the like.

In the electronic organizer configured as described above, when a voice is input to the microphone 2 while the button 1 is pressed, the voice is recognized. Then, the result of the voice recognition is displayed (output) on the display 3 and is output from the speaker 4 as a synthesized sound.

The user looks at the display on the display 3 and
Alternatively, if the user recognizes the output of the speaker 4 and the speech recognition result is incorrect, the user operates the button 1 again and speaks, for example, “cancel”. This "cancel" is registered as a command for instructing the state of the apparatus to return to the state before starting the processing when a certain process is performed in the electronic organizer. When a command “cancel” is received after performing a certain process, the process returns to a state before the process is performed (hereinafter, such a process in the electronic organizer is appropriately referred to as a cancel process).

Therefore, for example, as described above, when the speech recognition result displayed on the display 3 is incorrect, when the user speaks the command “cancel”,
The voice recognition result displayed on the display 3 is deleted,
The internal state of the device also returns to the state before performing voice recognition.

The canceling process is also performed by, for example, clicking (double-clicking) the button 1 twice consecutively. Therefore, in the electronic organizer, when the command “cancel” is input by voice or when the button 1 is double-clicked, the cancel process is performed.

The canceling process can be performed a plurality of times in succession. Therefore, if the command “cancel” is continuously input, or if the double click of the button 1 is continuously performed,
The state of the device sequentially returns to the previous state.

FIG. 2 shows the main functions of the electronic organizer of FIG. In this embodiment, the electronic organizer has a schedule management function and a directory management function (address book) function.
According to the schedule management function, the schedule management function stores the input date and time and the schedule in association with each other, and searches and outputs the stored schedule in response to a request from the user (display 3). Or output from the speaker 4). Further, according to the name list management function, the input customer and other names, the telephone number, and the address can be stored in association with each other, and the stored information can be searched and output in response to a request from the user. It has been done.

The electronic organizer further has an input function for inputting information to be stored and managed in the schedule management function and the name list management function, commands and the like. This input function mainly includes a voice recognition function,
There is a date / time relative designation function and a kana-kanji conversion function. According to the voice recognition function, voice recognition of the voice input to the microphone 2 enables input of information and other information by voice. . According to the date / time relative designation function, the relative date / time is input by voice, and when this is recognized by the voice recognition function, the relative date / time can be converted into an absolute date / time. Thus, when inputting or searching for a schedule, the date and time of the schedule for performing the input or search can be specified by a relative date and time. According to the kana-kanji conversion function, when the voice input to the microphone 2 is recognized, the character string as a result of the voice recognition can be converted to kana-kanji.

FIG. 3 shows an example of the internal configuration of the electronic organizer of FIG. The input unit 1 includes, for example, an A / D converter 2A in addition to the button 1, the microphone 2, and the cursor key 5. In the input unit 1, when the button 1 or the cursor key 5 is operated, a signal corresponding to the operation is supplied to the system control unit 12. In the input unit 1, an analog audio signal output from the microphone 2 is converted into a digital audio signal by A / D conversion in an A / D converter 2A. The audio signal output from the / D converter 2A is supplied to the system control unit 12 only while the button 1 is operated.

The input unit 1 can optionally include a keyboard having various keys, a pointing device, an OCR, and the like, as in the related art.

The system control unit 12 is, for example, a DSP (Di-
gital Signal Processor) and CPU (Central Processo)
r Unit) for controlling the entire apparatus. That is, for example, the system control unit 12
A signal from the input unit 12 is received, a process corresponding to the signal is performed, and data (information) for the storage unit 13 is stored.
Other writing and reading are performed. Further, the system control unit 12 supplies and outputs data to the output unit 14, for example, and performs a predetermined process based on the current date and time (date and time) supplied from the clock unit 15. I have. Also, the system control unit 1
2 is, for example, a voice recognition unit 16 or a voice synthesizer 17
By transmitting and receiving data to and from the respective devices, voice recognition or voice synthesis is performed respectively.

The storage unit 13 is, for example, a ROM (Read Only)
Memory), a RAM (Random Access Memory), a magnetic disk, an optical disk, an IC card, and the like, and stores, for example, a system program and an application program. The system control unit 12 performs various processes and controls by executing the system program and the application program stored in the storage unit 13. Also, the storage unit 1
Reference numeral 3 also stores a dictionary for performing kana-kanji conversion (hereinafter, appropriately referred to as a kana-kanji conversion dictionary), and the system control unit 12 refers to this and performs kana-kanji conversion. I have. Further, the storage unit 13 stores data necessary for the system control unit 12 to perform processing. The storage unit 13 also stores information (input data) such as a schedule input by the user and an address book.

The output unit 14 includes the display 3 and the speaker 4, an amplifier 4A and the like. Output unit 14
In this configuration, a signal (information) from the system control unit 12 is displayed on the display 3 or adjusted to an appropriate level by the amplifier 4A and then output from the speaker 4. The clock unit 15 generates a current date (date) and time (hereinafter, appropriately referred to as the current date and time) by counting a predetermined clock, and supplies the current date and time to the system control unit 12.

The voice recognition section 16 performs a voice recognition process. That is, when a voice signal is supplied from the input unit 1 to the system control unit 12, the system control unit 12 supplies the voice signal to the voice recognition unit 16, and the voice recognition unit 16 The audio signal from the system control unit 12 is acoustically analyzed, and speech recognition is performed based on the analysis result. This speech recognition result is converted into a corresponding character string, and the speech recognition unit 16
From the system controller 12.

The voice synthesizing section 17 performs a voice synthesizing process. That is, the system control unit 12 converts a character string (text) to be a synthesized sound, for example,
When the voice synthesizer 17 receives a character string from the system controller 12, the voice synthesizer 17 generates a synthesized sound (voice signal) corresponding to the character string and supplies it to the system controller 12. It has been made.

The system control unit 12 supplies the character synthesizing unit 17 with not only the character string but also information indicating the reading, accent position, phrase (intonation), etc. (hereinafter referred to as phoneme information as appropriate). The speech synthesizer 17 generates prosody information for controlling the prosody of the synthesized speech based on the phoneme information, and further performs system control based on the phoneme information and the prosody information. A synthetic sound corresponding to the character string from the unit 12 is generated. However, it is also possible to supply only a character string from the system control unit 12 to the speech synthesizing unit 17. In this case, the speech synthesizing unit 17 linguistically analyzes the character string, Phonetic information is generated based on the result.

Next, FIG. 4 shows an example of the configuration of the speech recognition section 16 of FIG. The audio signal supplied from the system control unit 12 is supplied to the acoustic analysis unit 21. The sound analysis unit 21 performs sound analysis of the sound signal from the system control unit 12 in units of a frame, which is a predetermined minute section, and thereby extracts the feature amount. That is, in the acoustic analysis unit 21, for example, the power (energy) of the audio signal, the number of zero crossings, the pitch frequency, the frequency characteristics,
Further, the amount of change (difference value) and the like are extracted as needed.

In order to obtain the frequency characteristics of the audio signal, for example, LPC analysis (linear prediction analysis), FFT
(Fast Fourier transform), filtering by a filter bank constituted by a band-pass filter, and the like are performed.

When the acoustic analysis unit 21 extracts the feature amount of the voice, it outputs this to the recognition unit 22, for example, as a vector, or as a scalar amount by vector quantization of the vector.

As described above, the speech feature amount (frame-based feature amount) is supplied to the recognition unit 22 as a time series.

The recognition unit 22 includes a recognition parameter storage unit 23
The voice input to the microphone 2 is recognized based on the feature amount from the acoustic analysis unit 21 while referring to the recognition dictionary storage unit 24 and, if necessary, the recognition grammar storage unit 25. A character string representing the recognition result is output.

Here, as a speech recognition algorithm for performing speech recognition in the recognition unit 22, for example, DP
(Dynamic Programming) matching method, method using neural network, HMM (Hidden Markov Mo
del) method.

In the DP matching method, a speech feature is used as an input pattern, and speech recognition is performed by matching the input pattern with a standard pattern called a template while performing expansion and contraction on the time axis. The character (character string) corresponding to the standard pattern that minimizes the distance from the input pattern is the speech recognition result.

A method using a neural network is as follows.
Speech recognition is performed using a network model that imitates the human brain.For example, when speech features are given to a network model, the speech recognition result is determined based on data output from the feature model. You.

The HMM method performs speech recognition using a probabilistic model (HMM) composed of several states. For example, a character (corresponding to a probabilistic model having the highest probability of observing a sequence of speech feature values) is used. (Character string) is the speech recognition result.

The recognition parameter storage unit 23 stores, for example, recognition parameters such as parameters defining a template, weight coefficients defining a network model, and probabilities defining a probability model. The recognition dictionary storage unit 24 stores a recognition dictionary in which speech recognition target words to be subjected to speech recognition in the recognition unit 22 are registered. The recognition grammar storage unit 25 stores a recognition grammar describing a grammar of a language to be subjected to speech recognition and a semantic connection relationship of a speech recognition target word registered in a recognition dictionary.

The recognition unit 22 calculates the likelihood (score) of the speech recognition target word registered in the recognition dictionary with respect to the sequence of the speech feature amount in one speech section, and the recognition parameter stored in the recognition parameter storage unit 23. The speech recognition target word having the highest score (likelihood) (or the top several words) may be obtained while restricting the recognition grammar stored in the recognition grammar storage unit 25 as necessary. Output as a speech recognition result.

Here, in the present embodiment, for example, the HMM method is used, and the speech recognition processing performed by the speech recognition unit 16 according to the HMM method will be described. Note that the speech recognition unit 16 can use a speech recognition algorithm other than the HMM method. Here, it is assumed that speech recognition is performed for Japanese, for example.

In this case, first, a probability model corresponding to, for example, a pseudonym is obtained by learning. That is, HM
M (probability model) is configured by connecting a plurality of states by paths, and the state transitions from one state to another state (including the original state) (state transition probability) and the state transitions. And the probability that each symbol is output (symbol output probability). In the learning, the state transition probability and the symbol output probability of the probability model corresponding to each kana are determined so that a predetermined evaluation value for the learning data of the voice of the kana is maximized.

More specifically, the state transition probability and the symbol output probability are obtained, for example, by observing a sequence of features (symbol sequence) obtained from data for learning the speech of the pseudonym in a probabilistic model corresponding to the pseudonym. Is determined so that the probability of occurrence (occurrence probability) is increased. The state transition probability and the symbol output probability of the probability model corresponding to each kana determined in this way are stored in the recognition parameter storage unit 23 as recognition parameters.

At the time of recognition, the sound analysis unit 2
In step 1, the input speech signal is subjected to acoustic analysis to be a feature amount of the speech, and the feature amount is vector-quantized to be a time-series symbol, which is supplied to the recognition unit 22. Is done.

Upon receiving the symbol, the recognizing unit 22 connects the probability models corresponding to the kana constituting the target words for speech recognition registered in the recognition dictionary, thereby forming a probability model corresponding to the target words for speech recognition. Generate. That is, when the speech recognition target word is, for example, “today (today)”, the probability models corresponding to the pseudonyms “to”, “to”, and “to” are connected, and the probability model corresponding to “today” is obtained. Generated. Further, when the speech recognition target word is, for example, “Asu (tomorrow)”, the probability models corresponding to the pseudonyms “A” and “SU” are connected to generate a probability model corresponding to “Asu”. Further, when the speech recognition target word is, for example, “ichi (1)”, the probability models corresponding to the pseudonyms “i” and “chi” are connected, and a probability model corresponding to “ichi” is generated. Further, when the speech recognition target word is, for example, “ni (2)”, the probability model of the pseudonym “ni” is used as the probability model of “ni” as it is. Furthermore, when the speech recognition target word is, for example, “san (3)”, the probability models corresponding to the pseudonyms “sa” and “n” are connected to generate a probability model corresponding to “san”.

The recognition unit 22 generates the probability models corresponding to all of the speech recognition target words registered in the recognition dictionary in this manner, and generates the probability transition models and the symbol output probabilities stored in the recognition parameter storage unit 23. Based on the probability model corresponding to each speech recognition target word, a sequence of speech feature amounts supplied from the acoustic analysis unit 21 is observed (occurred).
Is calculated. And the recognition unit 22
The highest score, that is, the character string representing the speech recognition target word corresponding to the probability model giving the occurrence probability is output as the speech recognition result together with the score.

In the present embodiment, a probabilistic model corresponding to a kana is prepared, and the probabilistic models corresponding to each kana are connected in accordance with a target word for speech recognition registered in a recognition dictionary, whereby the target word for speech recognition is obtained. Since the probability model is generated, for example, the recognition dictionary stored in the recognition dictionary storage unit 24 is replaced with another one, or a new recognition target word is added to the recognition dictionary. By simply deleting the words, the words to be recognized can be changed.

Further, for example, the numeral 4 or the like means “Yon”
May be uttered, or may be uttered as "shi". By registering both in the recognition dictionary, the voice 4 can be recognized regardless of which utterance is made. It becomes possible.

Further, generally, for example, words are often registered in the recognition dictionary as words to be subjected to speech recognition.
An arbitrary character string such as "from time to 10:00)" can be registered as a speech recognition target word. In this case, it is possible to perform voice recognition of any character string as described above in addition to words.

However, if too many words for speech recognition are registered in the recognition dictionary, as described above, the recognition performance is degraded, and the time required for the speech recognition processing is increased. Therefore, in the present embodiment, the recognition unit 22 refers to the recognition grammar stored in the recognition grammar storage unit 25 as necessary, and restricts the arrangement of the speech recognition target words based on the score ( Here, the occurrence probability is calculated as described above.

That is, for example, in the case of recognizing a voice having a sequence of time, “from”, time, and “to”, such as “from lottery to 100 (from 9:00 to 10:00)”, Numbers "0" to "2" used to represent time
4 ”,“ Ji (hour) ”,“ from ”and“ to ”are registered in the recognition dictionary as speech recognition target words, and the recognition grammar“ (number) ”+“ "+
"From" + "(number)" + "ji" + "to" are registered in the recognition grammar storage unit 25.

In this case, the recognizing unit 22 refers to not only the individual score of the speech recognition target words registered in the recognition dictionary but also the recognition grammar to obtain the score of the sequence of the speech recognition target words as described above. Is also calculated. Then, as a result, a character string corresponding to the highest score is output as a speech recognition result. Accordingly, the recognizing unit 22 is capable of performing voice recognition not only of the voices individually uttered for the voice recognition target words registered in the recognition dictionary, but also of the voices continuously uttering them. That is, the recognition unit 22 performs not only the isolated word recognition but also the continuous speech recognition.

Since the number of speech recognition target words and the number of recognition grammars that can be handled in a range where the recognition unit 22 can exhibit sufficient recognition performance are limited, the speech to be recognized is input. In this case, it is desirable that the number of speech recognition target words and the number of recognition grammars used for speech recognition of the speech be adaptively limited.

In the above case, the probability model (state transition probability and symbol output probability) corresponding to the pseudonym is stored in the recognition parameter storage unit 23. It is also possible to use phonemes, syllables, words, and the like.

Next, FIG. 5 shows the recognition dictionary storage unit 24 of FIG.
2 shows a configuration example of the recognition dictionary stored in the storage device. In this embodiment, the recognition dictionary is composed of a command dictionary, a one-letter word dictionary, a continuous character string dictionary, a basic dictionary, and a new word dictionary, and speech recognition target words are classified and registered in these dictionaries. (Registered).

In the command dictionary, commands such as "cancel" described above are registered. Further, in the command dictionary, when searching the schedule or the address book stored in the storage unit 13, the command dictionary includes words that indicate search items (for example, schedules, telephone numbers, addresses, and the like) to be searched. The instruction word is also stored.

That is, in the electronic organizer shown in FIG. 1, for example, a search for a schedule at a certain date and time is performed by the voice "(date and time)".
What is your schedule? "And the like. For example, the search for the address or telephone number of a certain person is performed by inputting a voice such as "What is the address of (person's name)?" Or "What is the telephone number of (person's name)?" ing. In this case, “schedule”, “address of”, and “phone number” in “schedule”, “address” and “phone number” are searched for schedule, address, Indicates a phone number. In the command dictionary, in addition to the commands, descriptive words (for example, "schedule is", "address is", "phone number is") indicating such a search item (search target) are also registered. Have been.

In the one-character word dictionary, a character (character string) representing one syllable and a character represented by one character (both are referred to as a one-character word, as appropriate) are registered. That is,
Specifically, as shown in FIG. 6, for example, as a character representing one syllable, “A”, “I”, “U”,.
, But also one character), “Kyu”, “Kyu”, “Kyo”,... Also,
As characters represented by one character, alphabets, numbers, symbols, and the like are registered.

A continuous character string is registered in the continuous character string dictionary. The continuous character string will be described later.

In the basic dictionary, words (system registration words) to be used for inputting, for example, a schedule or an address book are registered. That is, for example, when inputting (or searching for) a schedule, it is necessary to specify the date and time of the schedule. In the basic dictionary, a word for specifying such date and time (hereinafter, appropriately referred to as date and time) Etc.) are registered. Here, an example of the date designation word is shown in FIG. FIG. 7A shows a word (date designation word) for designating a date among date designation words.
FIG. 7B shows a word (time designation word) for designating time.

When the user sets words not registered in the basic dictionary as words to be subjected to speech recognition,
The word is registered. That is, when the user is, for example, a business person, words such as “meeting”, “business trip,“ negotiation ”,“ meeting ”, and“ departure ”frequently appear when entering a schedule. Used for
When the user is, for example, a student, words such as “class”, “English conversation”, and “piano” are frequently used when inputting a schedule. When such a word is not registered in the basic dictionary, the mode of the device can be input by voice by setting the mode of a kana-kanji conversion mode described later. Inputting in the kana-kanji conversion mode is not efficient. Therefore, such a word (hereinafter, appropriately referred to as a new word) can be registered in a new word dictionary, whereby the input of a word not registered in the basic dictionary can be performed by voice. It has been made so that it can be performed efficiently.

The speech recognition unit 16 performs speech recognition on the speech recognition target words registered in all of the command dictionary, the one-character word dictionary, the continuous character string dictionary, the basic dictionary, and the new word dictionary. Of these dictionaries, for example, for basic dictionaries and new word dictionaries, the words registered in them are classified into, for example, those used for inputting the date and time, location, business requirements, etc. At the same time, when inputting information (data) to the device, the user is required to specify what kind of information (information such as date and time, place, and business) to be input. Alternatively, the words classified as described above may be switched so as to be subjected to speech recognition.

Next, the system control unit 12 (FIG. 3) supplies the voice signal from the input unit 11 to the voice recognition unit 16 to perform voice recognition, and outputs the voice recognition result output from the voice recognition unit 16. The final speech recognition result of the speech uttered by the user is determined by the system control unit 12. Therefore, the determination process performed by the system control unit 12 will be described with reference to the flowchart of FIG.

When the system control unit 12 receives the speech recognition result and the score output from the speech recognition unit 16, the system control unit 12
In step S1, it is determined whether the score is equal to or greater than a predetermined value. If it is determined in step S1 that the score is not equal to or greater than the predetermined value, the process proceeds to step S2, where the system control unit 12 issues a warning by controlling the output unit 14, and ends the process.

That is, when the score of the speech recognition result is not equal to or more than the predetermined value, it is considered that the speech of a word not registered in the recognition dictionary has been input. Therefore, the system control unit 12
Causes the display 3 or the speaker 4 to output a warning message, thereby notifying the user that the input voice is not targeted for voice recognition.

On the other hand, if it is determined in step S1 that the score is equal to or greater than the predetermined value, the flow advances to step S3 to determine whether there are a plurality of voice recognition results output from the voice recognition unit 16. That is, the voice recognition unit 16
When multiple strings with the highest score are obtained (for example,
When the word is spoken, for example, when the same word is registered in a plurality of dictionaries constituting the recognition dictionary,
A character (character string) corresponding to the word registered in each of the plurality of dictionaries is obtained as a speech recognition result with the highest score), and the speech recognition results are output together with the score. In this case, in step S3, it is determined that there are a plurality of speech recognition results.

If it is determined in step S3 that a plurality of speech recognition results do not exist, that is, if the speech recognition result is one character string, the process proceeds to step S11, and the character string is determined as the final speech recognition result. It is determined, and the process ends.

If it is determined in step S3 that there are a plurality of speech recognition results, the process proceeds to step S4.
It is determined whether or not any of the plurality of speech recognition results has a command or instruction word registered in the command dictionary. If it is determined in step S4 that any of the plurality of speech recognition results has a command or instruction word registered in the command dictionary, the process proceeds to step S5, and the command or instruction A speech recognition result having a word is selected, and the process proceeds to step S11. In step S11,
The selected speech recognition result is determined as the final speech recognition result, and the process ends.

On the other hand, if it is determined in step S4 that none of the plurality of speech recognition results has a command or instruction word, the process proceeds to step S6, and the plurality of speech recognition results are determined. It is determined whether there is a one-letter word registered in the one-letter word dictionary. In step S6, when it is determined that one of the plurality of speech recognition results includes a one-letter word,
The process proceeds to step S7, the one-letter word is selected, and the process proceeds to step S11. In step S11, the selected one-letter word is determined as the final speech recognition result, and the process ends.

If it is determined in step S6 that there is no one-character word in any of the plurality of speech recognition results, the process proceeds to step S8, in which any of the plurality of speech recognition results includes It is determined whether there is a continuous character string registered in the continuous character string dictionary. If it is determined in step S8 that there is a continuous character string in any of the plurality of speech recognition results, the process proceeds to step S9, the continuous character string is selected, and the process proceeds to step S11. In step S11, the selected continuous character string is determined as the final speech recognition result, and the process ends.

On the other hand, if it is determined in step S8 that there is no continuous character string in any of the plurality of speech recognition results, that is, if the plurality of speech recognition results indicate a command, a descriptive word, a one-letter word, And neither a contiguous string,
Therefore, when the word is a system registered word registered in the basic dictionary, or a new word registered in the new word dictionary (hereinafter, referred to as a normal word, including both as appropriate) (or a character string composed of a normal word) , The process proceeds to step S10, and one of the plurality of ordinary words is selected.

That is, in this case, the system control unit 12
A plurality of ordinary words are displayed on the display 3 together with the cursor. Further, the system control unit 12 also causes the display 3 to display a message requesting selection of a normal word. The user sees this message and operates the cursor key 5 to move the cursor to the position of the normal word as a correct speech recognition result. Confirm word selection. In step S10, the normal word thus determined is selected.

When a normal word is selected in step S10, the process proceeds to step S11, where the selected normal word is determined as a final speech recognition result, and the process ends.

As described above, when a plurality of speech recognition results are obtained, the system control unit 12 sets the last priority in the order of the command or instruction word, one character word, continuous character string, and ordinary word. A typical speech recognition result is determined.

Note that the command (and descriptive word) is given the highest priority because when the user utters the command, the voice is recognized as a one-letter word, a continuous character string, or a normal word. If this happens, processing corresponding to the command will not be executed.

In this embodiment, the command registered in the command dictionary is recognized as a command for the apparatus. However, the command "cancel" as described above is registered as, for example, a normal word. In some cases,
When the voice “cancel” is input, the voice is always recognized as “cancel” of the command according to the above-described priority, and is not recognized as a normal word.
Therefore, in this case, using the word "cancel"
It becomes difficult to describe the schedule. Therefore, in order to prevent such a situation, when a command is input, for example, a command for instructing a process is input following a "command" or other voice that is not normally emitted. The voice input following the voice "command" can be recognized as representing the command.

Further, here, the priorities are given in the order of the command and the instruction word, the one-letter word, the continuous character string, and the ordinary word. However, the priorities can be given in another order.

Further, among the plurality of speech recognition results, there may be a plurality of commands, instruction words, one-letter words, or continuous character strings. In such a case, step S5
In S7 and S9, for example, as in the case of step S10, any one of the commands,
A descriptive word, a one-letter word, or a continuous character string is selected.

Further, when a normal word is obtained as the final speech recognition result, the system control unit 12 stores the normal word in the storage unit 13 (for example, if the kana-kanji conversion is possible). It is converted into a kanji by referring to it and handled in a kanji state (for example, the normal word is displayed in the kanji on the display 3).

Next, the processing performed by the system control unit 12 shown in FIG. 3 in response to the sound input to the microphone 2 will be described with reference to the flowchart shown in FIG. First, in step S21, the system control unit 12 determines whether or not sound has been input to the microphone 2. here,
The system control unit 12 determines whether or not a voice is input to the microphone 2 depending on whether or not the button 1 is operated. That is, when the button 1 is being operated, the system control unit 12
It is determined that a voice has been input to.

If it is determined in step S21 that no sound has been input to the microphone 2, the process proceeds to step S21.
Return to When it is determined in step S21 that a voice has been input to the microphone 2, the process proceeds to step S22, in which the system control unit 12 supplies the voice signal to the voice recognition unit 16 to perform voice recognition. afterwards,
When receiving the speech recognition result from the speech recognition unit 16, the system control unit 12 determines the final speech recognition result as described with reference to FIG. Then, proceeding to step S23, the system control unit 12 determines whether or not the speech recognition result is a command. If it is determined in step S23 that the voice recognition result is not a command, the process proceeds to step S24, where the system control unit 12 generates a message requesting the input of the command, and transmits the message to the voice synthesis unit 1.
7 to generate a synthesized sound corresponding to the message. Further, in step S24, the system control unit 12 supplies the corresponding synthesized sound to the output unit 14 together with the message and causes the output unit 14 to output the synthesized sound. That is,
As a result, a message requesting input of a command is displayed on the display 3, and the speaker 4 outputs a synthesized sound corresponding to the message.

After that, the process returns to step S21, and waits for the input of a voice, and the processes after step S22 are performed again.

On the other hand, if it is determined in step S23 that the speech recognition result is a command, the process proceeds to step S2.
Proceeding to step S5, the operation mode of the apparatus is set to a mode for performing a process corresponding to the command, and step S26 is performed.
Proceed to. In step S26, a process corresponding to the mode set in step S25 is performed, and thereafter, the process returns to step S25, and the processes in steps S25 and S26 are repeated.

The flowchart of FIG. 10 shows the details of the processing in step S26 when the mode of the apparatus is set to the schedule input mode for inputting a schedule. When the mode of the apparatus is set to the schedule input mode, in steps S31 to S33, the same processes as those in steps S21 to S23 in FIG. 9 are performed.

If it is determined in step S33 that the voice recognition result is a command, the schedule input mode is ended, that is, the process of step S26 in FIG. 9 is ended, and the process returns to step S25. Therefore, when it is determined in step S33 that the input voice is a command, in step S25, the operation mode of the apparatus is set in accordance with the command, and in step S26, the operation mode corresponding to the set mode is set. Is performed.

Here, when the mode of the apparatus is set by operating a key or the like, the mode of the apparatus is usually changed from the current mode to the upper (or highest) mode once. It is necessary to return to and select a desired mode. On the other hand, in the electronic organizer shown in FIG. 1 (FIG. 3), since the mode can be selected by voice, the mode of the apparatus can be changed directly from the current mode. That is, when the user switches the mode of the apparatus to the desired mode, the user does not need to once issue a command for returning to the higher mode, and further does not need to issue a command corresponding to the desired mode. , It is only necessary to say a command corresponding to a desired mode.

On the other hand, if it is determined in step S33 that the voice recognition result is not a command, the process proceeds to step S3
Proceeding to No. 4, it is determined whether or not the result of the voice recognition indicates the date and time. If it is determined in step S34 that the speech recognition result does not indicate the date and time, the process proceeds to step S35, and the user is requested to input the date and time in the same manner as in step S24 of FIG.
It returns to step S31.

That is, when a schedule is input, it cannot be managed unless the schedule is known. Therefore, the system control unit 12 requests input of the date and time of the schedule before inputting the schedule.

Here, the date and time of the schedule can be input by using a date and time designation word as shown in FIG. These date designation words can be used alone or in combination. That is, the input of the date and time is, for example, “today”
And "Tomorrow", "13:00", and further, such as "Next Monday", "1st of next month", and "1 am". The date and time can be input, for example, between “next week” and “Monday” or “next month”
It can also be performed by voice such as "Next Monday" with "no" inserted between "1" and "1st" or "1st of next month".

As shown in FIG. 7, there are various date and time designating words. Considering the case where a numeral representing the date and time or the case where the above "no" is inserted into the word, the method of expressing the date and time is as follows. , A huge number. If one of these huge numbers of expressions is selected and the date and time are to be input using the keyboard or a pointing device, the operation is rather complicated. Become.

On the other hand, in the case of inputting the date and time by voice, the date and time designation word as shown in FIG. 7 is registered in the recognition dictionary, and the connection rules such as the date and time designation word and the number are recognized by the recognition grammar. If it is described as, even if the date and time is uttered by any of the huge numbers described above, the voice can be recognized, and as a result, the input of the date and time can be easily performed. Can be. In other words, the user operates the button 1 while, for example, “January 1, 1995”
Alternatively, the user can input the date and time only by making an utterance such as “Next Monday”, “1:00 pm”, and “from 9:00 to 10:00”.

The time designation words (FIG. 7B) of the date / time designation words include, for example, “about hour”, “morning”,
There are vague expressions such as "day" and "night". For example, "about" means how much before and after the time, and "morning", "day", "night" Is set in advance from what time to what time.

If there is a voice input using the above-mentioned date and time designation word, that is, if there is a voice input of a date and time, it is determined in step S34 that the voice recognition result indicates the date and time. Then, the process proceeds to step S36.
In step S36, it is determined whether or not the date and time as the speech recognition result indicates a relative date and time. If it is determined in step S36 that the date and time do not represent a relative date and time, that is, if the date and time represents an absolute date and time, step S37 is skipped and step S37 is skipped.
Proceed to 38.

If it is determined in step S36 that the date and time represent a relative date and time, the process proceeds to step S36.
Proceeding to 37, the relative date and time is converted to an absolute date and time. That is, the system control unit 12 refers to the current date and time supplied from the clock unit 15 and converts the relative date and time into an absolute date and time.
Specifically, for example, when the date and time represented by the relative date and time are
If "Next Monday", the current date is 19
If it is Friday, March 1, 1995, the relative date and time “Next Monday” is converted to the absolute date and time “March 4, 1995”.

When the relative date and time are converted to the absolute date and time in step S37, the process proceeds to step S38, and a schedule input process is performed. That is, as in the case of step S35, the user is requested to input a schedule. In response to this, when the schedule is input by voice, the voice is recognized by voice and is associated with the absolute date and time. Can be The speech recognition result associated with the absolute date and time, that is, the schedule is supplied from the system control unit 12 to the storage unit 13 and stored. When the above-described schedule input processing is completed, the process returns to step S31, and the processing from step S31 is repeated again.

If the date and time entered by the user is only the time and does not include the date, the system control unit 12 determines that the time is, for example, the time on the current date. Be recognized.

By the way, when a schedule or the like is input by voice, and then the schedule is searched and displayed, if the display of the schedule is made only by a pseudonym, it becomes very difficult to see. Therefore, when inputting a schedule, it is desirable to convert a character string representing the schedule into kana-kanji characters.

Now, assuming that a schedule, for example, “in front of the west exit ticket gate of Shinagawa Station”, which represents a meeting place, is to be entered, to enter this, say “Shinagawa Eki no Nishiguchi Saisatsu”. In order to perform Kana-Kanji conversion by speech recognition, for example, the words “Shinagawa”, “Eki”, “No”, “Nishiguchi”, “Kai Satsu”, and “Mae” are used as speech recognition target words. It must be registered in the recognition dictionary and the fact that they are connected is registered in the recognition grammar storage unit 25 as a recognition grammar.

However, registering such words and recognition grammars so as to be able to handle all users leads to deterioration in recognition performance and reduction in recognition processing speed, as described above. become.

Therefore, here, for example, the kana-kanji conversion of a word that is not the target of speech recognition can be performed as follows.

That is, for example, when the character string “Shinagawa Station” is to be entered, this is entered as “shi”,
Speak one character at a time, such as "na", "ga", "wa", "e", "ki", "no" (hereinafter, such an utterance is appropriately referred to as one character utterance). One such character is
As described above, since the one-character word is registered in the one-character word dictionary (FIG. 6) constituting the recognition dictionary, the one-character utterance can be recognized by the voice recognition unit 16.

After the speech recognition of the one-character utterance,
A continuous character string is created by arranging a plurality of continuous characters in a character string obtained as a result of the speech recognition. That is, for example, “shi”, “na”, “ga”, “wa”,
For "e", "ki", and "no", a continuous character string "shina", "naga", "gawa", in which two consecutive characters are arranged,
"Waye", "Eki", "Kino" are created, and a continuous character string "Shinaga", "Nagawa", "Gawae", "Waeki""," Ekino "is created. Similarly, a continuous character string "Shinagawa", "Nagawae", "Gawaeki", "Waekino", which is a sequence of four consecutive characters, and a continuous character string "Shinagawae" in which five consecutive characters are arranged "," Nagawa Eki "," Gawa Ekino ", a continuous character string with six consecutive characters" Shinagawa Eki "," Nagawa Eki ", a continuous character string with seven consecutive characters" Create Shinagawa Eki no.

As described above, a continuous character string (here,
“Shina”, “Naga”, “Gawa”, “Way”, “Eki”,
"Kino", "Shinaga", "Nagawa", "Gawae", "Waeki", "Ekino", "Shinagawa", "Nagaue",
“Gawaeki”, “Wakino”, “Shinagawa”, “Nagaiwaki”, “Gawaeki”, “Shinagawaeki”, “Nagaiwaki”, and “Shinagawaeki” After the creation of 21 consecutive character strings), the continuous character string is used as a speech recognition target word, so that when a voice corresponding to the continuous character string is subsequently input, the voice is Be able to recognize.

As a result, for example, when “Eki” is not a speech recognition target word, “Shi”, “Na”, “Ga”, “Wa”, “E”, “K”,
When the voice "Eki" is input after the voice "No" is input, the continuous character string "Eki" is regarded as a target word for voice recognition. The voice is recognized by "Eki". If the speech recognition result is a continuous character string, the kana-kanji conversion is performed on the continuous character string.

As a result, for example, the voice “shi”,
When "na", "ga", "wa", "e", "ki", and "no" are input and then the voice "eki" is input, the character string consisting only of the kana "Shinagawa Eki""Eki" in "No" is converted to Kana-Kanji characters and becomes "Shinagawa Station".

As described above, by creating a continuous character string and using it as a new speech recognition target word, it is possible to specify a character string to be converted to kana-kanji by voice. That is, in this case, by simply registering the one-letter word in the recognition dictionary as described above, if it is a continuous character string created from characters that have been input once, its speech can be recognized. A character string to be converted into kanji can be specified by voice.

Further, in this case, when a continuous character string is uttered, the kana-kanji conversion is performed. Therefore, the utterance of the continuous character string means not only that the character string to be converted to kana-kanji is specified, but also that It also has the meaning as a trigger, which causes the kana-kanji conversion of a continuous character string.

Next, the operation of the system control unit 12 when performing the above-described kana-kanji conversion will be described with reference to the flowchart of FIG. When a command to perform kana-kanji conversion is issued by voice, the system control unit 1
2 indicates that the mode of the device is the kana-kanji conversion mode, and FIG.
The processing according to the flowchart shown in FIG. That is,
First, in steps S41 to S43, FIG.
0, the same processing as in each of steps S31 to S33 is performed.

If it is determined in step S43 that the voice recognition result is not a command, the process proceeds to step S43.
Proceeding to 44, it is determined whether the speech recognition result is a one-letter word. If it is determined in step S44 that the speech recognition result is a one-letter word, the system control unit 12
Causes the one-letter word to be displayed on the display 3, and in step S45, an unprocessed character string in which the one-letter word is added is created. The unprocessed character string will be described later.

Thereafter, in step S46, a continuous character string as described above is created for the unprocessed character string created in step S45. That is, step S46
In the case where the unprocessed character string consists of N characters, a continuous character string consisting of two consecutive characters, a continuous character string consisting of three consecutive characters in the unprocessed character string, ..., N consecutive characters Is created.

After the continuous character string is created, step S
Proceeding to 47, the continuous character string is registered in the continuous character string dictionary, whereby the continuous character string is set as a speech recognition target word.

Here, in the continuous character string dictionary, step S4
When the continuous character strings created in step 6 are sequentially added, the number of words for speech recognition increases, which affects the recognition performance and the recognition processing speed. Therefore, here, registration of the continuous character string created in step S46 in the continuous character string dictionary is performed by deleting the already registered continuous character string. By doing so, it is possible to prevent the continuous character strings registered in the continuous character string dictionary from increasing without limit.

The deletion of the continuous character string registered in the continuous character string dictionary is performed in step S47 in addition to the registration of a new continuous character string, for example, when ending the kana-kanji conversion mode, or It is also possible to execute the command when a predetermined command is input from the user.

After the continuous character string has been registered in the continuous character string dictionary as described above, the process returns from step S47 to S41. In step S41, after inputting a new voice, the process returns to step S42 and subsequent steps. Processing is performed.

On the other hand, if it is determined in step S44 that the speech recognition result is not a one-letter word, the process proceeds to step S44.
Proceeding to 48, it is determined whether the speech recognition result is a continuous character string registered in the continuous character string dictionary. If it is determined in step S48 that the speech recognition result is a continuous character string, the process proceeds to step S49, where the continuous character string is converted into kanji by referring to the kana-kanji conversion dictionary stored in the storage unit 13. Is done. Further, step S
At 49, the character portion constituting the continuous character string displayed on the display 3 is converted into the corresponding kanji,
Proceed to step S45.

In step S45, the character string following the character string (continuous character string) converted to kanji in step S49 is regarded as an unprocessed character string, and the processing in step S46 and thereafter is performed.

That is, for example, as described above,
"Shi", "na", "ga", "wa", "e", "ki",
When "no" is input to create a continuous character string and the continuous character string "Shinagawa" is uttered, it is converted to the kanji character "Shinagawa". In this case, in step S45,
The character string "ekino" following "Shinagawa" is regarded as an unprocessed character string. Then, in this state, when the continuous character string “Eki” is uttered, it is converted into “Station”,
In this case, in step S45, the character "no" following "station" is set as an unprocessed character string.

Thereafter, for example, the one-letter words "ni", "shi", "gu", "chi", "ka", "i",
If “sa”, “tsu”, “ma”, and “e” are sequentially uttered, in step S45, the unprocessed character string “no”
Then, the uttered one-letter word is sequentially added, and an unprocessed character string “Nishi-Nishiguchi-kaisatsu-mae” is created.

In this state, for example, when "Nishi" is uttered, "Nishi" is a continuous character string. The kanji is converted to the kanji “west”, and the unprocessed character string is changed to “guchikaisatsumae”. Less than,
The continuous character strings "Guchi", "Kai-satsu", and "Mae" are sequentially uttered, and these are converted to Kanji, and finally, a mixed sentence of Kanji kana "Shinagawa station west exit ticket gate" is obtained. Can be

In addition, as described above, the one-letter word "shi",
After uttering "na", "ga", "wa", "e", "ki", "no", and further uttering the continuous character strings "shinagawa" and "eki", the one-letter word "ni" , "Shi", "gu",
"Chi", "ka", "i", "sa", "tsu", "ma",
If "e" is spoken and the continuous character strings "nishi", "guchi", "kaisatsu", and "mae" are spoken,
The display on the display 3 is "Shinagawa Ekino", "Shinagawa Ekino", "Shinagawa Station's", "Shinagawa Station's Nishiguchi Satsumasae", "Shinagawa Station Nishi Gachikai Satsuma", "Shinagawa Station's It changes to "West exit ticket gate", "Shinagawa station west gate ticket gate", "Shinagawa station west gate ticket gate". The display 3 can display the unprocessed character string portion, for example, in an inverted manner. In this case, the user can clearly recognize the portion that is currently the unprocessed character string.

On the other hand, if it is determined in step S48 that the speech recognition result is not a continuous character string, the process proceeds to step S48.
Proceeding to 50, a request is made to the user to speak a one-character word or a continuous character string, and the process returns to step S41.

As described above, a continuous character string is created from an unprocessed character string, and is used as a speech recognition target word. Thus, even if a word is not registered in advance as a speech recognition target word, speech recognition can be performed. It is possible to perform the kana-kanji conversion.

As a kana-kanji conversion dictionary for performing kana-kanji conversion, a dictionary with a large vocabulary has already been realized, and such a kana-kanji conversion can be used here. . As a result, it is possible to handle such a large-scale kana-kanji conversion dictionary by voice without using all words registered therein as speech recognition target words in advance.

Next, the flowchart of FIG. 12 shows the details of the processing in step S26 in FIG. 9 when the operation mode of the apparatus is set to the new word mode for registering a new word. A new word that is not registered in the basic dictionary can be input by performing one-character utterance as described above, and can further perform its kana-kanji conversion. However, if the new word is frequently used, It is troublesome to have to utter one character each time such a new word is input.

Therefore, in the present embodiment, such a new word can be registered in the new word dictionary by using the above-described kana-kanji conversion.

When registering a new word, the user inputs a command for registering a new word by voice. In this case, the system control unit 12 sets the mode of the apparatus to the new word registration mode, and performs processing according to the flowchart shown in FIG. That is, in steps S61 to S70, the system control unit 12 executes step S41 in FIG.
The same processing as in steps S50 to S50 is performed.

However, in step S69 corresponding to step S49 in FIG. 11, the uttered continuous character string is registered as a new word in the new word dictionary.

Therefore, the user can register a new word to be registered by inputting the new word by one-character utterance and further uttering the new word.

In step S69, not only the uttered continuous character string (new word) is registered in the new word dictionary, but also the kana-kanji conversion of the continuous character string is performed.
This can be registered in the new word dictionary in association with a continuous character string as a new word.

In this case, when the user utters the new word, the kanji associated with the speech recognition result can be immediately displayed.

That is, as a new word, for example, “Kaite Kyoshi” is associated with “Katekyo”, which is converted into a kanji, and registered in the new word dictionary. When “Kyoshi” is input, the kanji “tutor” associated with the recognition result of the voice can be displayed immediately.

Also, if new words are allowed to be registered without restriction,
Since the voice recognition performance and the recognition processing speed are reduced, it is preferable to notify the user when the number of registered new words reaches a predetermined number.

Next, the processing in step S26 in FIG. 9 when the mode of the apparatus is set to the search mode for searching the schedule and the address book will be described with reference to the flowchart in FIG. Here, it is assumed that, for example, the above-mentioned "schedule of", "address of", and "phone number of" are registered as command words in the command dictionary. However, the designation word is not limited to these.

In this case, first, steps S81 to S8
3, the processing is performed in the same manner as in steps S31 to S33 in FIG.

If it is determined in step S83 that the voice recognition result is not a command, the process proceeds to step S83.
Proceeding to 84, it is determined whether the speech recognition result includes the instruction word. If it is determined in step S84 that the speech recognition result does not include the instruction word, the process proceeds to step S85, and the user is instructed to specify the search item in the same manner as in step S24 of FIG. Is requested, and the process returns to step S81.

On the other hand, if it is determined in step S84 that the speech recognition result includes the descriptive word, the flow advances to step S86 to determine the word placed before the descriptive word (hereinafter, appropriately referred to as a prefix). Is searched for the search item represented by the descriptive word.

That is, for example, when the voice “What is the schedule for January 1, 1995?” Is input, the prefix “199” arranged before the instructive word “schedule of” is input.
A search for information on "January 1, 5" is performed on a search item "schedule" represented by the directive word "schedule." Therefore, in this case, in step S86,
The schedule of January 1, 1995 is retrieved from the storage unit 13.

For example, when the voice “(person name) is the telephone number?” Is inputted, the information search for the prefix “(person name)” is performed, The search item "telephone number" is searched. Therefore, in this case, in step S86, the telephone number of (person name) is retrieved from the storage unit 13. Further, for example, when the voice “address of (person name)” is input, the search for information about the prefix “(person name)” is performed, and the search item “address of Is performed on the subject. Therefore, in this case, in step S86, the address of (person name) is retrieved from the storage unit 13.

After the information search is performed in step S86, the process proceeds to step S87, in which the search result in step S86 is output from the output unit 14 (displayed on the display 3 or output from the speaker 4). It returns to S81.

In the above-described information search, speech recognition of the date and time is possible as described with reference to FIG. 10, but speech recognition of a personal name is not always possible. Since there is no personal name,
It is desirable to register as a new word as described in.

As described above, according to the electronic organizer of FIG. 1, information and other inputs are made by voice, so that the input can be made easily without increasing the size of the apparatus. Furthermore, it is possible to input information and other flexible inputs, for example, relative dates and times, and words that are not previously subjected to voice recognition.

The case where the present invention is applied to an electronic organizer has been described above. However, the present invention is applicable to any device that inputs data by voice.

In this embodiment, 1-letter word dictionary contains 1
Although only the character words are registered, an arbitrary word can be registered in the one-character word dictionary.

Further, in this embodiment, a continuous character string is created from the speech recognition result of one-character utterance, and this is used as a new speech recognition target word. A continuous character string is created from a character string output from 4 and representing information other than the speech recognition result,
This can be newly set as a speech recognition target word. That is, for example, in a case where the present invention is applied to an electronic dictionary device or the like, when a sentence describing the word is displayed as a search result of a word corresponding to the input voice, continuous characters from the sentence are displayed. It is possible to create a column. In this case, even if a word constituting a sentence is not previously set as a speech recognition target word, the word can be speech-recognized. Therefore, when the word is uttered, it is necessary to search for the word. Becomes possible.

Further, in this embodiment, the language to be subjected to speech recognition is Japanese, but the language to be subjected to speech recognition is not limited to Japanese.

[0174]

The speech recognition method and the speech recognition apparatus according to the present invention.
According to the location , it was not previously targeted for speech recognition
Even a word can be recognized by speech .

[0175]

[0176]

[Brief description of the drawings]

FIG. 1 is a perspective view showing a configuration of an embodiment of an electronic organizer to which the present invention is applied.

FIG. 2 is a diagram showing main functions of the electronic organizer of FIG. 1;

FIG. 3 is a block diagram showing an example of an electrical configuration of the electronic organizer of FIG. 1;

FIG. 4 is a block diagram illustrating a configuration example of a speech recognition unit 16 in FIG. 3;

FIG. 5 is a diagram illustrating a configuration of a recognition dictionary stored in the recognition dictionary of FIG. 4;

FIG. 6 is a diagram showing registered contents of the one-letter word dictionary of FIG. 5;

FIG. 7 is a diagram showing registered contents of a basic dictionary of FIG. 5;

FIG. 8 is a flowchart illustrating a method for determining a final speech recognition result.

FIG. 9 is a flowchart illustrating a process of the system control unit 12 of FIG. 3;

FIG. 10 is a flowchart illustrating details of a process in step S26 in FIG. 9;

FIG. 11 is a flowchart illustrating details of a kana-kanji conversion process.

FIG. 12 is a flowchart illustrating details of a process in step S26 in FIG. 9;

FIG. 13 is a flowchart illustrating details of a process in step S26 in FIG. 9;

[Explanation of symbols]

1 button, 2 microphones, 3 display, 4
Speaker, 11 input section, 12 system control section,
13 memory unit, 14 output unit, 15 clock unit, 1
6 speech recognition unit, 17 speech synthesis unit, 21 sound analysis unit, 22 recognition unit, 23 recognition parameter storage unit, 24 recognition dictionary storage unit, 25 recognition grammar storage unit

Continued on the front page (51) Int.Cl. ⁷ identification code FI G10L 3/00 561E (58) Fields investigated (Int.Cl. ⁷ , DB name) G10L 15/00-15/28 G06F 17/22

Claims (5)

(57) [Claims] 1. A speech recognition method for recognizing an inputted speech by inputting a character string consisting of one character word, and arranging a plurality of characters continuous from the character string consisting of one character word. a creation step of creating a continuous string, targeting the continuous string, the speech recognition method characterized by comprising a speech recognizing speech recognition step. The speech wherein inputted, a speech recognizing method for recognizing speech with reference to the recognition dictionary speech recognition terms that are the object of speech recognition is registered, the string of character words An inputting step of inputting, a creating step of arranging a plurality of continuous characters in the character string composed of the one-letter word to create a continuous character string, and registering the continuous character string in the recognition dictionary. And a registration step as the speech recognition target word. 3. A display step of displaying a character string composed of the one-letter word, and when the voice is recognized by the continuous character string registered in the recognition dictionary, the displayed character string includes sound
3. A kana-kanji conversion step of converting a kana-kanji conversion of a voice-recognized continuous character string.
Voice recognition method described in. 4. In the recognition dictionary, a character represented by at least one character is registered in advance as the speech recognition target word, and a character string composed of the one-character word is one character.
4. The speech recognition method according to claim 3, wherein the speech recognition is performed by speech recognition of a word . 5. A speech recognition apparatus for recognizing an input speech, comprising: a storage unit for storing a recognition dictionary in which speech recognition target words targeted for speech recognition are registered ; Input means for inputting a character string consisting of, in the character string consisting of the one- letter word, to create a continuous character string by arranging a plurality of consecutive characters, the continuous character string,
A speech recognition apparatus, comprising: a registration unit that registers the speech recognition target word in the recognition dictionary.