JP2647873B2 - Writing system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (a) Field of Industrial Use The present invention relates to an apparatus for creating a sentence by voice recognition.

(B) Conventional technology Sound is recorded on a recording / playback device such as a tape recorder,
A sentence creation system for performing speech recognition by inputting a playback sound output by reproducing the speech to a speech recognition device and converting the sentence into a sentence has been developed (Japanese Patent Laid-Open No. 58-1587).
No. 36).

2. Description of the Related Art Conventionally, in a device such as a sentence creating system in which a recording / reproducing device, a voice synthesizing device, a voice recognizing device, and a display device are combined, a portion for displaying a voice reproduced from the recording / reproducing device and data read from a storage device on a display device. And the reading part of the speech synthesis operated independently.

In such a device, for example, when an error occurs during confirmation of a prepared sentence or the like, and the recording / reproducing device wants to hear a portion corresponding to the error portion, it is necessary to listen to the sentence recorded in the recording / reproducing device from the beginning. Had to find the relevant part.

(C) Problems to be Solved by the Invention Conventionally, when a sentence created by a word processor or the like is read by a speech synthesis function, an error in the sentence is found, and a portion corresponding to the error portion is found from a recording / reproducing device, I had to play back and listen to the recorded audio, or use the cue function to find out what was wrong. In each case,
After finding an error in the sentence using the speech synthesis reading function,
The recording / reproducing device required a complicated operation to find the part corresponding to the erroneous part and to reproduce the part.

In view of the above, the present invention makes it possible to execute a complicated operation to find a part corresponding to an erroneous part from a recording / reproducing apparatus and to reproduce the same by an extremely simple operation after finding an error in a sentence by a reading function of speech synthesis. It is to be.

(D) Means for Solving the Problems The sentence preparation system of the present invention uses a silent section recorded in a recording / reproducing apparatus as a unit of recognition unit, so that the input voice reproduced from the recording / reproducing apparatus is used as a syllable or syllable. Has a voice recognition function for recognizing each word, and when a silent section is detected, a silent section detection signal indicating that a silent section has been detected is recorded in the silent section, and the result of voice recognition is recognized for each recognition unit. A sentence to be stored in a storage device with a delimiter signal added thereto, and a silent section detection signal and a delimiter symbol are made to correspond to each other on a one-to-one basis. And a sentence composition for synchronizing the display output and the speech synthesis output between the sentence speech synthesized by the speech synthesis unit corresponding to the result of the speech recognition and the sentence speech recorded in the recording / reproducing device. In the system, the recording / reproducing device has a function of reproducing a sentence voice synchronized with the voice-synthesized portion, and when the synthesized sound is stopped, for example, when correcting the speech recognition result, the reproducing portion of the recording / reproducing device has a predetermined function. It is characterized in that it returns backward by the silent section detection signal.

(E) Function According to the system of the present invention, when configured in this way, for example, an error is found during reading by speech synthesis,
When the synthesizing function is stopped, the recording / reproducing apparatus returns from the current position by a predetermined silent section detection signal and stops. In other words, at this time, the recording / reproducing apparatus returns by a predetermined silence section detection signal from the text synthesized by the synthesizing unit, and thus returns to a part before the part corresponding to the part where the voice synthesis output is stopped. The error part has been found.

(F) Embodiment FIG. 1 shows an external view of a dictating machine which employs the present invention to create sentences by voice input, and FIG. 2 shows a functional block diagram of the machine.

In FIG. 2, (1) is a speech recognition unit provided in the main body (100) of FIG. 1, and details thereof are shown in a block diagram of FIG. Preprocessing unit (11) [FIG. 4], and a feature extracting unit (12) [FIG. 5] for extracting a parameter indicating the acoustic feature from the sound pressure adjusted sound signal from the processing unit (11). , The extraction unit (12)
Word recognition unit (13) that performs word recognition of input speech based on feature parameters obtained from
4) [Fig. 7] and any of these recognition units (13),
A candidate creating unit (15) for creating a candidate for a recognized word character string or a recognized syllable character based on the recognition result from (14).

Further, in FIG. 2, (2) is a recording / reproducing device such as a tape recorder which can be mechanically and electrically detachably attached to the main body (100) as shown in FIG. 1, and (3) is, for example, the one shown in FIG. A headphone type microphone (4) is an input switching unit [8th] for switching connection between the recording / reproducing device (2), the microphone (3) and the voice recognition unit (1).
Figure]. (6) a display device for displaying a character string or the like generated based on a recognition result, (7) a keyboard for inputting various control signals of the dictating machine, and (8) a keyboard for inputting the dictating machine. A storage device such as a magnetic disk device for storing the character string generated by
(9) is a voice synthesizing unit for reading out a character string in the storage device from the speaker (10) by rule synthesis. still,
(5) is a control unit composed of a microprocessor, which controls the operation of each unit.

There are two methods for creating a sentence using the dictating machine having the above-described configuration, and each will be described in detail below.

The first method is to input a raw voice from a microphone (3) to a voice recognition unit (1), perform voice recognition, convert the input voice to a character string, display the same on a display device (6), and simultaneously store the storage device. The result is stored in (8).

A second method is to record a sentence to be input in a recording / reproducing device (2) in advance, connect the recording / reproducing device (2) to the present device, and input the recorded sentence to a voice recognition unit (1). Performs voice recognition, converts the input voice into a character string, displays it on the display device (6), and simultaneously stores the result in the storage device (8).

As described above, there are two methods for inputting voice, and the input switching unit (4) switches the input. In addition to the input switching, the input switching unit (4) also switches between recording the recording signal (a) in the recording / reproducing device (2) and recording the voice input from the microphone (3). .

Hereinafter, the operations from voice recording to text creation will be sequentially described in detail.

(I) Voice Registration Processing Before performing voice recognition, voice registration is performed to create a standard voice pattern required for voice recognition.

 First, the syllable registration mode will be described.

The standard pattern described here is a reference pattern at the time of pattern matching in the phrase recognition unit (14) of the speech recognition unit (1). Specifically, the standard pattern as shown in FIG. 14) Syllable standard pattern memory (14
stored in d).

To register a voice to this dictating machine,
First, the switch (14s1) in FIG. 7 is operated to connect the parameter buffer (14a) to the syllable standard pattern memory (14a), and there are three methods described below.

The first method is to place a microphone (3) on the body (100) of the machine.
A method of directly inputting a registered voice, analyzing the registered voice in a voice recognition unit (1), creating a standard pattern, and storing the created standard pattern in a syllable standard pattern memory (14d) and a storage device (8) It is.

The second method is to connect the recording / reproducing apparatus (2), which has previously recorded the registered voice, to the main body (100) and reproduce the recorded voice, thereby inputting the registered voice, and inputting the registered voice. Is analyzed by the voice recognition unit (1), a standard pattern is created, and the created standard pattern is stored in the syllable standard pattern memory (14d) and the storage device (8).

The third method is to use a microphone (3) on the main unit (100) of the machine.
The registered voice is input directly from the main unit (100). At this time, the recording / reproducing device (2) is connected to the main unit (100), and the input voice is recorded on the recording / reproducing device (2). Then, the registered voice from the microphone (3) is analyzed to create a standard pattern, and the created standard pattern is stored in the storage device (8). Then, when the voice input to the microphone (3) is completed,
The voice recorded by the recording / reproducing device (2) is reproduced, the registered voice recorded is analyzed by the voice recognition unit (1), a standard pattern is created, and the created standard pattern is stored in a syllable standard pattern memory (14d). At the same time as the syllable standard pattern of the registered voice directly from the microphone (3).

In the third method, the sound recorded in the recording / reproducing device (2) receives the frequency characteristics of the recording / reproducing device (2), so that the standard pattern created from the recorded sound and the microphone (3) When a standard pattern created from a voice directly input from a PC is compared, a difference appears between the two standard patterns. Therefore, when recognizing the recorded voice, it is necessary to use a standard pattern created from the recorded voice, and when recognizing the voice directly input from the microphone (3), the standard pattern created from the voice directly input from the microphone (3) is used. Since it is necessary to use the standard pattern, the above-described method is used to create both the standard pattern directly registered from the microphone (3) and the standard pattern created from the recorded voice by one voice registration operation. You can remember. Also, once the registered voice is recorded in the recording / reproducing device (2), the registrant does not need to input the voice on the dictating machine on which the standard pattern has not been created. Thus, a standard pattern can be created.
Also, if the registered voice is recorded in the recording / reproducing device (2), and a sentence is recorded after the registered voice, the recording / reproducing device (2) is connected to the main body (100) later, and the recorded voice is recorded. Just by playing the sound, from voice registration to text creation,
Everything can be done automatically.

The registrant who creates the standard voice pattern is input by the registrant reading out characters displayed on the display device (6) in a certain order.

When the recording / reproducing device (2) having a dedicated display function of the machine is used, even when the recording / reproducing device (2) is carried alone, a voice corresponding to the headword displayed on the display screen is produced. By recording on the recording / reproducing device (2), a standard pattern can be created.

As described above, when a registered voice for creating a standard pattern is recorded in the recording / reproducing device (2), the recorded voice is affected by noise when creating the standard pattern from the recorded voice. There is a possibility that the headword may be different from the corresponding headword. Hereinafter, a case where a tape recorder is used as a recording / reproducing apparatus will be described with reference to FIG. FIG. 9A shows a state in which the registered voice for creating the standard pattern is recorded on the tape recorder, and the registered voice “A” corresponding to the headwords “A” to “KA” is shown.
This represents the state of the tape between "?" And "?", And here shows the case where [noise] is recorded between "e" and "o". As shown in FIG. 9 (a), when a voice is registered using a tape on which [noise] is recorded between the registered voices and the registered voice, the first recorded voice is "A" and the second voice is recorded. If the input sound is determined by the order of the sounds recorded on the tape, such as “I”, which syllable the input sound corresponds to, the registered sound will be up to [Noise]. Since the headword is regarded as corresponding, the input registered voice and the headword are shifted.

Here, in FIG. 9 (b), [noise] is erroneously recognized as speech, and [noise] is input at the headword "e".
FIG. 11 is a diagram in which a syllable “e” is input at the headword “o”.

As described above, when the standard pattern is created from the registered voice, the recorded voice may be displaced from the headword under the influence of noise or the like. Therefore, as shown in FIG. 9C, the type of the registered voice is indicated. The character code sound is recorded in the recording / reproducing device (2) in correspondence with the registered voice. With this method, even if [noise] is recorded between "u" and "e", a shift between the input sound and the headword is prevented as described above.

A method of recording a character code sound of a specific frequency for preventing this deviation will be described separately for a case where the tape recorder of the recording / reproducing apparatus (2) is a single track and a case where the tape recorder is a multi-track.

First, a case where a recording / reproducing apparatus having a multitrack is used as a recording method will be described with reference to FIG.

When a recording / reproducing apparatus having a multi-track is used as a recording method, a character code corresponding to a headword is recorded on a track on which no voice is recorded as shown in FIG. In the voice recognition unit (1), the headword of the input voice is known from the character code sound, and among the sounds recorded in the voice track, the sound recorded in the section t1 in which the character code sound was recorded is recorded. Of these, only those satisfying the condition equal to or higher than the sound pressure threshold value are regarded as speech and analyzed.

Also, as shown in FIG. 3B, a character code corresponding to the headword is recorded at the beginning and end of the voice, and a character code sound indicating the beginning of the voice among the sounds recorded on the voice track; Of the sounds recorded in the section t2 between the character code sounds indicating the end of the voice, only those that meet the condition of the sound pressure threshold or more are regarded as voice and analyzed.

Alternatively, a character code corresponding to the headword is recorded at the beginning of the voice, as shown in FIG. Of the sounds recorded in the voice track, from the character code sound indicating the type of this sound to the character code sound corresponding to the next headword, in the sound recorded in section t3,
Only those satisfying the condition equal to or higher than the sound pressure threshold value are regarded as speech and analyzed.

As a second method, in the case of a single-track recording / reproducing device (2), a character code corresponding to a headword is represented by a sound outside the analysis frequency band of the voice, and is recorded together with the voice on the track where the voice is recorded. . The method of recording the character chord sound in this case is the same as in the case of the multi-track described above. That is, t1, t described above
Of the sounds recorded in the section between t2 and t3, only those meeting the same conditions as above are regarded as voices and analyzed. However, except for the case of the embodiment shown in FIG. 10A in which the voice and the character code sound overlap, it is not necessary to use a sound outside the analysis frequency band of the voice as the character code sound.

Next, a word dictionary (13) of the word recognition unit (13) as shown in FIG.
The word standard pattern corresponding to the word registered in d) is stored in the word standard pattern memory (13
Register in c).

First, by a predetermined operation, the parameter buffer (13a) and the word standard pattern memory (13c) of FIG. 6 are connected by the switch (13s1), and the mode is set to the word registration mode.

Next, alphabets, numbers, parentheses, punctuation marks, and the like are displayed on the display device (6) of the apparatus main body (100), and the operator speaks the corresponding reading.

The voice recognition unit (1) analyzes this voice and registers the word standard pattern in the word standard pattern memory (13c).

Voice recognition becomes possible by the operations described above. However, the independent and auxiliary word dictionary (14e) and the word dictionary (13d)
If you want to recognize words that do not exist in the dictionary, register the words you want to recognize in the independent word /
It is necessary to register the word to be recognized in 3d) and the word standard pattern in the word standard pattern memory (13c).
However, register words in the independent word / attached word dictionary (14e), or use the word dictionary (13d) and word standard pattern memory (1e).
3c), whether to register words and word standard patterns,
The determination is made depending on whether the user wants the word to be recognized as a phrase utterance or a word utterance.

If the word is in the independent word / attached word dictionary (14e) but is not in the word dictionary (13d), but still wants to be recognized by word recognition, the word is stored in the word dictionary (13d) and the word standard pattern memory (13c). , Words and word standard patterns need to be registered.

 The method of registering an arbitrary word is described below.

To register a word, use the dictionary of independent words and attached words (14e)
There are two methods of registering a character string in the word dictionary, registering a word standard pattern in the word standard pattern memory (13c), and registering a character string in the word dictionary (13d).

When a word is registered in the independent word / attached word dictionary (14e), the word to be registered is uttered and input to the apparatus.

At this time, the apparatus recognizes the voice by the voice recognition unit (1) and displays the recognition result on the display device (6). If the result is correct, the user presses a predetermined key of the keyboard (7) and registers the uttered voice in the independent word / attached word dictionary (14e) as a character string displayed on the display device (6). if,
If the recognition result displayed on the display device (6) is not correct, the syllable correction function of the present device corrects the recognition result displayed on the display device (6), or re-utters the word to be registered. If the result of the re-utterance is incorrect, it is corrected again by the syllable correction function of the present apparatus. The above operation is repeated until the character string displayed on the display device (6) matches the word to be registered.

When the word is registered in the word standard pattern memory (13c) and the word dictionary (13d), first, the display device (6) is used in the same manner as when the word is registered in the independent word / attached word dictionary (14e).
Display the character string you want to register correctly. Next, the correctly recognized character strings and word standard patterns are stored in a word dictionary (13
d) and the word standard pattern memory (13c).

In addition, it is impossible to recognize a voice input with a natural utterance, considering the current level of voice recognition technology. At the current level of speech recognition technology, continuous speech utterance input is the limit, so one embodiment of continuous syllable utterance input will be described below.

In the case of continuous vocal utterance input, the above procedure is the same, but in the case of continuous syllable vocal input, the word to be registered is re-uttered by natural utterance because the word standard pattern is also a pattern of continuous syllable utterance. Then, a word standard pattern is created from natural utterance, and the word standard pattern and the character string are registered in the word standard pattern memory (13c) and the word dictionary (13d), respectively.

By the above operation, data necessary for creating a sentence by voice recognition can be registered.

(Ii) Text preparation An example of text preparation is described below.

First, when performing the recognition operation, the switch (13s1) of the word recognition unit (13) is connected to the switch (14s1) of the phrase recognition unit (14) such that the parameter buffer (13a) and the word determination unit (13b) are connected. ) Is set to connect the parameter buffer (14a) and the syllable recognition unit (14b).

 There are two ways to create sentences.

The first method is an online recognition method in which a sentence to be created in the main body of the apparatus is directly input by voice from a microphone (3).

The second method is offline recognition in which a recording / reproducing device (2) that has recorded a sentence is connected to the present device, and the recorded sentence is reproduced and recognized.

 First, an embodiment of online recognition will be described.

In the case of online recognition, a sentence uttered in units of phrases or words is directly input to the apparatus by voice from the microphone (3), and the microphone (3) and the voice recognition unit are input by the input switching unit (4) by a predetermined operation. Connect (1).

To record the sound input from the microphone (3) in the recording / reproducing apparatus (2), the recording / reproducing apparatus (2) is connected to the main body, and the input switching section (4) is connected to the microphone (3). ) Is connected to the recording terminal of the recording / reproducing device (2).

At the same time, the input switching unit (4) functions to record a beep indicating a phrase or a word break when a silence detection signal is input from the feature extraction unit (12) as described later.

At the time of speech recognition, word recognition unit (13) and phrase recognition unit (14)
Is running.

The voice input from the microphone (3) is sent to the pre-processing unit (11)
The input voice is processed so as to have characteristics suitable for voice analysis (for example, when the sound pressure of the input voice is low, a process of amplifying the sound pressure by an amplifier is performed), and a feature extraction unit (12) Sent to

In the feature extraction unit (12), as shown in FIG. 5, the speech input from the pre-processing unit (11) is analyzed by the analysis unit (12a) to perform feature extraction, and stored in the parameter buffer (12c).

At the same time, the analysis unit determination unit (12b) of the feature extraction unit (12)
Then, based on the analysis result of the analysis unit (12a), silence sections after being uttered in units of syllables or phrases, and beep sounds recorded after being uttered in units of phrases or words (for details, see In this case, a silent section detection signal (b) is generated when a silent section is detected.

The parameter buffer (12c) that has received the silent section detection signal (b) sends the stored feature parameters to the word recognition section (13) and the phrase recognition section (14), and erases the stored contents.

The feature parameters input to the word recognition unit (13) are stored in the parameter buffer (13a) shown in FIG. In the word determination unit (13b), the parameter buffer (13
A) The feature parameter stored in A) is compared with the word standard parameter memory (13c), and the parameter buffer (13a) is compared.
A word having a feature parameter and a word standard pattern having a large likelihood is selected from the word dictionary (13d), and the character string of the selected word and its likelihood value are sent to the candidate creation unit (15). send.

On the other hand, the feature parameters input to the phrase recognition unit (14) are stored in the parameter buffer (14a). In the syllable recognition unit (14b), the feature parameters stored in the parameter buffer (14a) and the syllable standard parameter memory (14
d), and converts the characteristic parameter stored in the parameter buffer (14a) into a syllable string, and sends the syllable string to the syllable determination unit (14c). The phrase determining unit (14c) compares the input syllable string with the words registered in the independent word / adjunct dictionary (14e), and combines the independent word and the adjunct word to generate a plurality of sets of phrases having a high likelihood. Then, the character string of the created phrase and its likelihood value are sent to the candidate creating unit (15).

The candidate creator (15) selects a plurality of candidates having a large likelihood from the input character string, and receives the likelihood value and the data sent from the word recognizer (13) or sent from the phrase recognizer (14). A code indicating the data is added and stored. At the same time, a signal for causing the display device to display a character string having the highest likelihood is sent to the control unit (5). The control unit (5) receives this signal, and adds a delimiter mark after the character string of the one having the highest likelihood. To the input sentence shown in FIG.
The image is displayed on the display device in a format as shown in FIG. At the same time, the candidate creating section (15) sends a signal to the control section (5) to store the content stored in the candidate creating section (15) in the storage device (8). The control unit (5) receives this signal, and stores it in the storage device (8) in a form in which a code representing a delimiter signal is added after the character string stored in the candidate creation unit (15). The character string stored in the external storage device is a primary document of a word processor. Generally, a floppy disk is used, but at this time, the format of the file in the storage device (8) needs to match the file format of a word processor.

The signal generating section (42) of the input switching section (4) shown in FIG. 8 which receives the silent section detection signal generates a beep sound indicating a passage of a sentence or a break of a word, and switches the beep sound. Enter in (41). The switch (41) connects a circuit so as to record the sound input from the microphone (3) and the beep sound input from the signal uttering unit (42) to the recording / reproducing device (2). A beep sound is recorded in a silent section that is regarded as a phrase or word segment of the sentence recorded in (2).

 Next, an embodiment of off-line recognition will be described.

In the case of off-line recognition, a sentence is created by reproducing and inputting a recorded voice of the recording / reproducing device (2) to the present apparatus. First, a sentence is recorded on the recording / reproducing device (2).

Further, in order to input a voice from the recording / reproducing device (2), the input switching unit (4) uses the recording / reproducing device (2).
And the voice recognition unit (1).

At the time of recording a sentence, utterance is made in units of phrases or words, and a silent section is created between the phrases and words. When a dedicated recording / reproducing apparatus (2) as shown in FIG. 1 is used, a beep sound indicating a delimiter is recorded on the recording / reproducing apparatus (2) or the book to clarify the delimitation of phrases and words. Press the delimiter key (71) set on the disk tating machine and record.

The registered words are uttered in word units. If the recording / reproducing device (2) has a character sound generating function and has a character corresponding to the word to be input, the character sound is replaced with the character sound. May be recorded.

In addition, in order to prevent the head recorded in each sentence or the noise recorded between sentences from being erroneously recognized as speech, a signal indicating the beginning and end of the sentence is recorded together with the speech.

However, the recording method of this signal is as follows:
Changes as described in the voice registration depending on whether or not is a multi-track system. FIG. 11 is a diagram of a multi-track system and FIG. 12 is a diagram of a single-track system. Eleventh
FIGS. 12 (a) and 12 (a) show a method of detecting a section in which a sound such as a DTMF signal is recorded as a voice area.

FIG. 11 (b) shows a sound such as a DTMF signal recorded before the start of the sentence, and FIG.
This is a method of re-recording and detecting a section sandwiched between the two signals as a voice area.

Further, in the case of the single track system shown in FIG. 12, a DTMF signal or the like out of the audio band is used, considering that the sound of the voice section and the sound of the DTMF signal and the like overlap.

Also, when recognizing a sentence, the section between 14 and t5 before and after the signal is recorded is sampled, and it is necessary to sample the beginning of the sentence and the beginning of the signal, and the end of the sentence and the end of the signal in order to determine whether or not it is a voice. It does not need to match. For this reason, it is possible to recognize even if the timing of uttering a sentence and the timing of pressing a key slightly deviate.

Next, the recording / reproducing apparatus (2) is connected to the main body of the present apparatus to reproduce the recorded voice and perform the recognition process. Before recognizing the recorded voice, the mode of the recognition speed is changed to a higher reproducing speed of the recorded voice. The fast-recognition mode for shortening the recognition time is either the mode for recognizing at the normal playback speed or the double-playback recognition mode when there is sufficient time and a high recognition rate is required. Set the mode.

 First, an embodiment of the early listening recognition mode will be described.

In the fast-recognition recognition mode, the playback speed of the recorded voice is increased, so the characteristics of the input voice are different from those of the standard pattern created from the registered voice played at the normal playback speed. Even if you input a voice that speeds up
Cannot perform accurate speech recognition.

Therefore, the sampling frequency is changed in order to accurately recognize the sound whose reproduction speed has been increased. Hereinafter, examples of such a method will be described.

The sampling frequency control unit (12d) of the feature extraction unit (12) in FIG. 5 sets the sampling frequency of the input voice of the feature extraction unit (12) to (the playback speed / recording speed) Speed) times,
Sample and analyze audio. The processing after the feature extraction unit (12) is the same as in the embodiment at the time of online recognition. However, in the recorded text of the recording / reproducing device (2), a sentence in which a peep sound indicating a delimiter for clarifying a segment and a word is input is input, and the feature extractor (12) detects the peep sound. Then, the feature extraction unit (12) generates a peep sound detection signal (b ') instead of the silent section detection signal (b). When the received signal is not a silent section detection signal (b) but a peep sound detection signal (b '), the signal generation section (42) of the input switching section (4) indicates a segment of a sentence or a word segment. No beep sound is generated.

When the voice recognition unit (1) recognizes a character sound indicating a word, it outputs a word corresponding to the character sound as a recognition result.

 Next, an embodiment of the double reproduction recognition mode will be described.

In this mode, first, a recorded voice is reproduced and input to the apparatus. At this time, the preprocessing section (11) of the voice recognition section (1) reads all the sound pressure fluctuations of the recorded voice, and stores this data in the sound pressure fluctuation memory (11b) shown in FIG. Next, the recorded voice is reproduced again and input to the apparatus. At this time, the pre-processing unit (11) uses the data stored in the sound pressure fluctuation memory (11b) and adjusts the input sound pressure to the feature extraction unit (12) as shown in FIG. The gain of the AGC circuit (11a) is adjusted to match the level. That is, the gain G
And multiplied by the fixed gain A of the control voltage V _G (it is variably adjusted).

Further, as another embodiment of the double playback recognition mode, a multiple playback recognition mode is also conceivable. This means that a recorded sentence is reproduced and input many times, and each time the input is input, the result recognized by changing the recognition method in the voice recognition unit (1) is compared, and the one with the highest likelihood of certainty is selected. How to

In addition, the recording / reproducing device (2) does not record the registration voice, and the recording / reproducing device (2) has different frequency characteristics when the reproduction speed is increased from the frequency characteristics when the reproduction speed is normal. Or when using a recording / reproducing device (2) having a frequency characteristic different from that of the recording / reproducing device (2) used for creating a standard voice pattern, or for creating a standard voice pattern. The recording / reproducing device (2) has the same frequency characteristics as the recording / reproducing device (2) in standard, but the actual frequency characteristics are different from those of the recording / reproducing device (2) used for creating the standard pattern of the voice due to the influence of errors of the parts used. When the apparatus (2) recognizes the recorded text, a function for correcting the influence of the frequency characteristic described below is used.

First, a reference sine wave signal serving as a reference when measuring the frequency characteristics of the recording / reproducing apparatus (2) is supplied to a reference signal generator (42).
And record it in the recording / playback device (2). The reference sine wave signal recorded after that is reproduced and input to the apparatus. The voice recognition unit (1) analyzes the input reference sine wave signal and obtains a difference between the frequency characteristics of the recorded reference sine wave signal and the reference sine wave signal generated by the reference signal generation unit (42). The correction is performed so as to reduce the difference in frequency characteristics between the recorded reference sine wave signal and the reference sine wave signal generated by the reference signal generator (42). A number of means for performing the correction can be considered by the feature extraction method of the feature extraction unit (12) of the speech recognition unit (1). For example, as shown in FIG. 13, if an analog filter bank system including a parallel connection of a band-pass filter (BPF) and an amplifier (AMP) connected in series is used, an amplifier (AMP) By adjusting the amplification factor, the output from the filter is adjusted so as to reduce the difference in the frequency characteristics from the reference sine wave signal generated by the reference signal generator (42). Also, if a digital filter is used as the feature extraction method of the feature extraction unit (12), the parameters that determine the characteristics of the digital filter may be changed. In addition, various methods are conceivable corresponding to the feature extraction method of the feature extraction unit (12) of the speech recognition unit (1).

By the operations up to the above, the sentence input by voice is changed to the kana sequence. A method of correcting a sentence in which the kana column changed is different from the input sentence will be described below with reference to FIG. 14 for each error type. Make corrections according to the following procedure.

14 (a) is an input sentence, FIG. 14 (b) is an input voice, FIG. 14 (c) is a recognition result, and FIGS. 14 (d) to (h) are correction processes.
FIG. 3I shows the result of the correction.

First, a correction method when a word spoken as a word is incorrectly recognized as a phrase will be described. When the word uttered as the word "C" is recognized as the phrase "shi" as shown in FIG. 3C, the cursor (X) is first moved to an incorrect word portion [FIG. ) I]. Next, the next word candidate key (72) is pressed to display the next word candidate [FIG. (D) ii]. If the result is correct, proceed to the next correction. If this result is incorrect, the next word candidate key (72) is pressed again to display the next word candidate. This operation is repeated until a correct answer is displayed.

Next, a description will be given of a correction method when a phrase uttered as a phrase is incorrectly recognized as a word. When the phrase uttered as "i" is recognized as the word "E", the cursor (X) is first moved to the wrong phrase. Next, the next phrase candidate key (73) is pressed to display the next phrase candidate. If the result is correct, proceed to the next correction.

If this result is incorrect, the next phrase candidate key (73)
Press to display the next candidate of the phrase. This operation is repeated until a correct answer is displayed.

By pressing the pre-word candidate key (74), a word can be displayed, and by pressing the pre-phrase candidate key (75), the phrase and the preceding candidate can be displayed.

When a correct answer cannot be obtained by the above two correction methods, correction is performed in units of syllables, and a word, a phrase, or a syllable is re-voiced.

Also, when re-utterance input, phrases are recognized again,
In order to avoid phrase recognition, the candidate creation unit (15) regards only the recognition result sent from the word recognition unit (13) as the recognition result and sends it from the phrase recognition unit (14). The recognition result can be externally controlled to be ignored.

In addition, the candidate creation unit (15) regards only the recognition result sent from the phrase recognition unit (14) as a recognition result, and ignores the recognition result sent from the word recognition unit (13) so as to ignore it. Can control.

The above-mentioned next candidate key is a key having the functions described below, and will be described with reference to FIG.

As described above, the speech recognition unit (1) of the present apparatus performs word recognition and phrase recognition in parallel, and seeks both word and phrase recognition results. The key for displaying the recognition results on the display device (6) in order from the one with the highest likelihood is the next phrase candidate key (73), and the result of the word recognition processing is displayed in the order from the one with the highest likelihood. The key for displaying the recognition result on the display device is the next word candidate key (72), and the recognition result currently displayed on the display device is good, and the recognition result with one large likelihood is displayed on the display device (6). The keys to be displayed are the pre-word candidate key and the pre-phrase candidate key.

FIG. 15 shows a candidate buffer (15a) of the candidate creating section (15). In this figure, the recognition result of the first place is "Tango", which is represented by (word) indicating that this is the recognition result sent from the word recognition section (13). Similarly, the second-ranked recognition result is "Tango", which is a phrase recognition unit (1
The recognition result sent from (4) is represented by (phrase), and the third-ranked recognition result is "Tango", which is the recognition result sent from the phrase recognition unit (14). Something is expressed by (clause), and the recognition result of the fourth place is "Tanko"
This is represented by (word) indicating that this is the recognition result sent from the word recognition unit (13).

Now, it is assumed that "Tango" is displayed on the display device (6). When the next phrase candidate key (73) is pressed in such a state, "Tango" is displayed on the display device (6). When the next word candidate key (72) is depressed, "Tanko" is displayed on the display device (6).

In addition, when "Tanko" is displayed on the display device (6), when the previous word candidate key (74) is pressed, "Tango" is displayed on the display device (6), and the pre-phrase candidate is displayed. Key (7
When 3) is pressed, "Tango ni" is displayed on the display device (6).

 Next, the batch correction method for an entire phrase is described.

The example of FIG. 14 (e) is an example in which the word “T” is erroneously recognized as “A”. First, the cursor is moved to the word to be corrected [FIG.

Next, the next word candidate key (72) is pressed to display the next word candidate [FIG. (E) ii]. If the result is correct, proceed to the next correction. If this result is incorrect, the next word candidate key (72) is pressed to display the next word candidate. This operation is repeated until a correct answer is displayed. If no correct answer is displayed, re-speak and re-enter. By pressing the previous word candidate key (74), the word candidate displayed immediately before can also be displayed.

 Next, a batch correction method for an entire word will be described.

The example of FIG. 14 (f) is an example in which the phrase “gamenno” is erroneously recognized as “gainenno”. First, the cursor is moved to the phrase to be corrected [FIG.

Next, the next phrase candidate key (73) is pressed to display the next phrase candidate [FIG. (F) ii]. If the result is correct, proceed to the next correction. If this result is incorrect, the next phrase candidate key (73) is pressed to display the next phrase candidate. Repeat this operation until the correct answer is displayed. If no correct answer is displayed, re-speak and re-enter. By pressing the previous phrase candidate key (75), the phrase candidate displayed immediately before can also be displayed.

 Next, a method for correcting syllable units will be described.

The example in FIG. 14 (h) is an example in which the phrase “onsende” is erroneously recognized as “onkeide”. In this example, the syllable “ke” is corrected to “se”. First, the cursor (X) is moved to the syllable “ke” to be corrected [(h) i in FIG. Press. By pressing the syllable next candidate key (76), the syllable closest in distance to the syllable to be corrected is displayed [FIG. (H) ii]. If the correct answer is displayed,
Move to the next revision. If this result is incorrect, the next syllable candidate key is pressed again to display the next syllable candidate. This operation is repeated until a correct answer is displayed. If the correct answer is not displayed, re-input is performed by re-utterance. If the result of re-entry is incorrect, correct it again by the above procedure. This operation is repeated until a correct answer is displayed.

By pressing the previous syllable candidate key (77), the candidate immediately before the syllable can be displayed.

To delete a syllable, move the cursor to the syllable you want to modify and press the delete key (78) to delete it.

To insert a syllable, move the cursor to the syllable you want to modify and press the insert key (79) to insert it.

 Next, the method of correcting several syllables will be described with reference to FIG.

This example is an example in which the input sentence "KAIJO" in FIG. 7A is erroneously recognized as "KANJO" in FIG. In this case, the cursor (X) is first brought to the syllable to be corrected (FIG. 3 (c)), and "kai" is entered again. The re-uttered input voice is recognized by the voice recognition unit (1), and the recognition result is displayed on the display device (6). If the recognition result is correct, the process proceeds to the next correction unit. If "kai" is erroneously recognized as "kae" as shown in FIG. 3D, if the word is a word, the next word candidate key (72) is pressed. If it is a phrase, press the next phrase candidate key (73). FIG. 16 shows an example of the case of a word, and a method of correcting the word will be described below. Figure (d)
If you press the next word candidate key (72) in the state of,
From the word dictionary (13d), the control unit (5) compares the recognition result “kanjo” in FIG. 12B before correction with the recognition result “kaejo” in FIG. Find the same part "Jo". Next, the control unit (5) selects a word having the same part "jo" from the word dictionary (13d). FIG. 11F shows the stored contents of the word dictionary (13d), and FIG. 13G shows the “Jo” selected from the stored contents.
Is shown. Next, the control unit (5) calculates the likelihood between the word described in FIG. 7G and the recognition result “Kaejo” after the re-speaking, and displays the word having the highest likelihood value. Figure (e)].

Next, a case of correcting a recognition boundary error of a phrase or a word will be described.

In the example of Fig. 14 (g), the phrase "bunsho-o" is between "n" and "shi". This is an example in which a silence section indicated by a mark is erroneously recognized, and is erroneously recognized in two parts, such as a word “bun” and a phrase “showo”. In this case, the recognition boundary error must be corrected. However, if it is desired to delete the recognition boundary delimiter, move the cursor (X) to the recognition boundary delimiter to be deleted [(g) i in FIG. Press [] (g) i
i]. If you want to insert a recognition boundary separator, move the cursor (X) to the syllable at the position where you want to insert it and press the insert key (79).

However, as described later, the beep sound of the recording / reproducing device (2) and the delimiter added to the recognition result stored in the storage device (8) are different from those of the recording / reproducing device (2) and the storage device (8). It must be taken into account as it serves as a marker for synchronization. Therefore, at this time, the fact that the delimiter is inserted or deleted in the storage device (8) is stored in the storage device (8).

For example, part of the sentence shown in FIG.
It is assumed that the data is stored in the storage device (8) as shown in FIG. (G) When the chapter of i is modified as shown in (g) ii, the delimiter stored in the storage device (8) Is changed to the symbol "" as shown in FIG. The symbol "" is a delimiter Is a symbol indicating that the symbol is deleted, and is not used as a symbol indicating a recognition unit, but is used only for control with the recording / reproducing device (2) or the like.

According to such a configuration, the delimiter Is deleted, the beep sound recorded in the recording / reproducing device (2) and the symbol stored in the storage device (8) By using "", both devices can be controlled while maintaining synchronization.

The above is a separator Is deleted, but the same idea can be applied to the case where the data is inserted. In other words, a specific symbol that is not used as a control signal but represents only a delimiter Should be inserted instead of

By the above correction procedure, as shown in FIG.
Correct the sentence.

After the recognition boundary error correction, the recognition unit in which the recognition boundary error correction is performed is corrected according to the correction procedure. In the case of correction by re-utterance, anyone who has registered the standard pattern can recognize any voice, so that the correction operation can be performed even if it is not the recorder of sentences.

The method of correcting the kana sequence has been described above, and the following functions are provided as functions for assisting the correction.

By moving the cursor on the character string displayed on the display device (6) and scrolling the display screen, the storage device (8) can sequentially display the stored sentences on the display screen. At this time, the sentences are displayed on the screen. Sound corresponding to the portion is reproduced from the recording / reproducing device (2).

It also has a function opposite to the above-mentioned function, and a character string corresponding to a portion being reproduced from the recording / reproducing device (2) is displayed on the display device (6).

In both of the above-mentioned methods, the delimiter sound recorded in the recorded text and the delimiter recorded on the display side are used as timing signals for synchronizing, and the reproduction of the recording / reproducing apparatus (2) is performed. And the display are controlled to operate in synchronization with each other. When a signal for stopping the reproduction is input from the keyboard (7) or the recording / reproducing device (2), the reproduction is stopped and the scrolling of the display or the movement of the cursor is stopped.

By the function of synchronizing the reproduction and display of the recording / reproducing apparatus (2) as described above, the character string can be recognized while listening to the reproduced sound, thereby facilitating finding a correction position.

As a method of synchronization described here, a method of displaying a character string of the storage device (8) corresponding to the part being reproduced on the display device (6), and a delimiter from the part corresponding to the part being reproduced. There is a method of displaying the kana row of the one-delayed portion on the display device (6).

In this case, when the display is stopped for correction, the corrected portion of the recorded voice has already been reproduced, and thus, to reproduce the corrected portion again, it is necessary to start the portion to be corrected from the reproduced text. Therefore, when this method is adopted, a function of automatically backtracking the recording / reproducing apparatus (2) to the immediately preceding delimiter when the display is stopped is provided.

When a tape recorder is used for the recording / reproducing device (2), the reproducing portion is controlled by the rotation of the motor, and when the cue of the portion corresponding to the corrected portion cannot be accurately performed due to slack of the tape or the like. There is.

In such a case, the input voice is stored for a certain period of time in PCM recording or ADPCM recording, and if you want to hear back the input voice, add a function to hear back the PCM recording or ADPCM recording voice. .

FIG. 17 shows an embodiment of the above function, and is a diagram of a PCM data memory for storing PCM recording data. Numerals 01 to 05 in the figure indicate addresses. The input voice is
“Iwa ｜ ten ｜ shi ｜ ear ｜ te ｜
This is a sentence that says "Gamen | Bunsho | Ten | Onsei | Onsen | Maru".

When the above sound is input, the PCM data memory (D
M) stores the voice “Iwa” up to the first silent section at address 01. At the address 02, the voice "ten" up to the second silent section is stored. At the address 05, the voice "te" up to the fifth silent section is stored. At this time, the PCM address pointer (AP) stores the address of the data stored first in the data stored in the PCM data memory. In this example, 01 is stored.

 At this stage, the PCM data memory is full.

Next, when voice is input, the PCM data memory (D
The input voice is stored at the address of the data stored first in the data stored in M). In this example, “gamenno” is stored at the address 01 where “Iwa” was stored. At this time, the PCM address pointer (AP) stores the address of the data stored first in the data stored in the PCM data memory (DM). In this example, 02 is stored.

In this state, when reproducing the contents of the PCM data memory (DM), reproduction is started from the address indicated by the PCM address pointer (AP). In this example, playback is performed in the order of 02, 03, 04, 05, 01.

By such a method, it is possible to hear the sound accurately and quickly many times.

Also, by moving the cursor (X) to the recognition unit delimiter signal on the screen and pressing the recording voice cue key (70), the recording / reproducing device (2) corresponding to the recognition unit indicated by the cursor is moved. Has a function of searching for a delimiter sound portion of a recorded sentence, and reproducing the sentence following it. An example of such a function will be described below.

To confirm the recognized text, the recognition result is read from the storage device (8) and displayed on the display device (6) from the beginning. At this time, the delimiter counter (5a) of the controller (5) in FIG. 19 counts the number of delimiters read from the symbol device (8). If the read recognition result is incorrect, place the cursor on the incorrect part and press the cue key. The control unit (5) reproduces the text recorded in the recording / reproducing device (2) in a fast-forward reproduction mode. Beeper counter (12e) of the feature extractor (12)
Counts beep sounds indicating a break in a sentence input from the recording / reproducing device (2).

The comparison circuit (5b) determines that the value of the beep counter (12e) is greater than the value of the delimiter counter (5a) described above.
When it is reduced by one, the signal (c) is sent to the recording / reproducing device (2), and the reproduction is stopped.

In addition, in order to confirm the recognition result and the sentence after the correction, the stored data of the storage device (8) is displayed as a character string on the display device (6), and the character string displayed on the display screen is visually checked. I have to read and read, which makes my eyes very tired.

In view of this point, the present apparatus has a function of reading out a character string on the storage device (8) storing the recognition result by a speech synthesis function, thereby recognizing the recognition result and the sentence after the correction is completed. This can be done by listening to the synthesized speech.

Also in this case, a delimiter is used as a timing signal for synchronizing the voice synthesizer (9), the storage device (8), the recording / reproducing device (2), and the display device (6).

In other words, a character string corresponding to the portion read by the voice synthesizer (9) from the storage device (8) is displayed on the display device (6), and at the same time, the recording portion is caught by the recording / reproducing device (2). . According to this method, when an error is found by the speech synthesis reading function and the speech synthesis reading function is stopped for correction, both the display on the display device (6) and the recording portion of the recording / reproducing device (2) have an error portion. Indicates that
Corrections can be made immediately.

As a method of synchronization described here, a kana column in a storage device corresponding to a portion read out by the speech synthesis function is displayed on the display device (6), and at the same time, a sentence recorded on the recording / reproducing device (2). A method of reproducing the corresponding syllable part and a method of reproducing the sentence part recorded in the recording / reproducing apparatus (2) one delimiter behind the part corresponding to the part read out by the speech synthesis function. is there. In the latter case, when the speech synthesis is stopped for correction, the recording / reproducing device (2) is stopped before the portion to be corrected, so that if the sound is reproduced in this state, the voice of the corrected portion can be reproduced immediately. In the former case, when speech synthesis for correction is stopped, the corrected portion of the recorded voice has already been reproduced, so that it is necessary to backtrack to reproduce the corrected portion again. Therefore, when the former method is adopted, when the display is stopped, the recording / reproducing device (2) is automatically
Preferably has the function of backtracking to the previous delimiter.

As described above, the embodiment in which the recognition result is stored in the storage connection (8) is described. However, as another embodiment, the recognition result may be stored in the recording / reproducing device (2).

If the recognition result stored in the storage device (8) is recorded in the recording / reproducing device (2) where the original is recorded, the original and the recognition result can be recorded on the same recording medium. Management becomes easier.

Also, by recording the recognized result in the recording / reproducing device (2) while reproducing and inputting the recorded text, an external storage device becomes unnecessary.

In any case, by using the multi-track recording and reproducing device (2), while reproducing the recorded voice,
The recognition result can be stored in a track on which no sound is recorded.

(G) Effects of the Invention According to the sentence creation system of the present invention, when an error is found while checking the created sentence and the synthesizing function is stopped, the playback portion of the recording / playback device returns by a predetermined silent section detection signal. As a result, the recording / reproducing apparatus has found the beginning of the erroneous part, and if the reproduction is performed in this state, the text at the erroneous part can be immediately heard from the recording / reproducing apparatus.

As described above, the original sentence (the sentence recorded on the recording / reproducing device) is simultaneously searched at the time of confirming the result of the speech recognition sentence using the speech synthesis function. Can be easily reproduced, and the efficiency of the correction operation can be improved.

[Brief description of the drawings]

FIG. 1 is an external view of a dictating machine employing the speech recognition system of the present invention, FIG. 2 is a structural diagram of the dictating machine, FIG. 3 is a configuration diagram of a speech recognition unit (1), FIG. Is a block diagram of the preprocessing unit (11), FIG. 5 is a block diagram of the feature extraction unit (12), FIG. 6 is a block diagram of the word recognition unit (13), and FIG. 7 is a block diagram of the phrase recognition unit (14). FIG. 8 is a structural diagram of the input switching unit (4), FIG. 9 is a diagram showing a relationship between a headword, a recording method, and a character sound, FIG. 10 is a diagram showing a relationship between a character sound recording method and a voice section, FIG. 11 is a diagram showing a recording method when the recording / reproducing apparatus is a multi-track system, and FIG.
FIG. 13 is a diagram showing a recording method when the recording / reproducing apparatus is a single track system, FIG. 13 is a diagram showing an example of a frequency correction circuit,
FIG. 14 is a correction diagram at the time of misrecognition, FIG. 15 is a diagram showing the candidate buffer (15a) in the candidate creating section (15), FIG. 16 is a diagram showing an example of correcting several syllables at the time of misrecognition, FIG. Is an illustration of PCM recording method,
FIG. 18 is an explanatory diagram of the AGC operation, and FIG. 19 is an explanatory diagram of a separator counter. (1) ... voice recognition unit, (2) ... recording and playback device,
(3) ... microphone, (6) ... display device, (7) ... keyboard, (8) ... storage device, (11) ... preprocessing unit,
(12): Feature extraction unit, (13) Word recognition unit, (14)
...... Phrase recognition unit.

Continuation of the front page (56) References JP-A-62-113264 (JP, A) JP-A-59-62949 (JP, A) JP-A-59-127148 (JP, A) JP-A-58-217044 (JP, A) , A)

Claims (1)

(57) [Claims] 1. A speech recognition function for recognizing an input voice reproduced from a recording / reproducing device for each syllable, syllable, or word unit by using a silent section recorded in the recording / reproducing device as a delimiter of a recognition unit. When a silent section is detected, a silent section detection signal indicating that the silent section has been detected is recorded in the silent section, and the speech recognition result is stored in a storage device with a delimiter for each recognition unit. A detection signal and a delimiter signal are made to correspond one-to-one, a sentence to be displayed on a display device corresponding to the result of the voice recognition using the two signals, and a voice synthesis unit corresponding to the result of the voice recognition. In a sentence creation system that synchronizes the display output and the speech synthesis output between the sentence speech synthesized by the And the function of reproducing the sentence voice from the recording / reproducing device, and when the synthesized sound is stopped, for example, when correcting the speech recognition result, the reproducing portion of the recording / reproducing device reverts by a predetermined silent section detection signal. Sentence writing system.