JPH01106096A - Voice recognition system - Google Patents

Abstract

PURPOSE: To shorten a recognition time by recognizing a speech reproduced while the reproduction speed of a sound recording and reproducing device is made faster than the speed at the time of sound recording. CONSTITUTION: In quick hearing recognition mode, the reproduction speed of a sound-recorded speech is made fast, so a sampling frequency is varied so as to accurately recognize the speech which is made fast in reproduction speed. Namely, a sampling frequency control part 12d of a feature extraction part 12 sets the sampling frequency of the input speech of the feature extraction part 12 (reproduction speed/sound recording speed) time as high as the sampling frequency when a standard pattern of the speech is generated and the speech is sampled and analyzed. Thus, the sampling frequency is varied to obtain a speech pattern of good precision, there is no deterioration in recognition rate, and the recognition time can be shortened.

Description

DETAILED DESCRIPTION OF THE INVENTION (a) Industrial field of application The present invention relates to a speech recognition system that creates sentences by recognizing speech that is reproduced and input from a recording and reproducing device.

(b) Conventional technology A voice system in which a clear voice is recorded on a recording/playback device such as a tape recorder, and the reproduced voice is input into a voice recognition device to perform voice recognition and convert it into text. A recognition system is being developed (Japanese Unexamined Patent Application Publication No. 58-15
No. 8736).

In such conventional systems, when recognizing the voice recorded on the recording/playback device, the voice recorded on the recording/playback device is played back at the same speed as when recording the voice.

(c) Problems to be Solved by the Invention Conventionally, the voice recorded in a recording/playback device was played back at the same speed as when the voice was recorded, so the same amount of recognition time as the time of recording was required.

In view of this point, the present invention makes it possible to shorten the time required to recognize speech recorded in a recording/playback device.

(d) Means for Solving the Problems In the speech recognition system of the present invention, the playback speed of the recording and playback device is made faster than the speed at which the recording was performed, and the reproduced speech is input to the recognition device.

At this time, for example, the recognition device uses
Import audio by doubling the sampling frequency (playback speed/recording speed).

(e) Effects According to the system of the present invention, the recognition time can be shortened because the voice reproduced by the recording and reproducing device at a higher reproduction speed than the recording speed is recognized.

Embodiment FIG. 1 shows an external view of a desk cheating machine which employs the present invention and creates sentences by voice input, and FIG. 2 shows a functional block diagram of the machine.

In Fig. 2, (1) is a speech recognition section equipped with a circuit in the main body (100) of Fig. 1, and its details are shown in the block diagram of Fig. 3. A pre-processing section (11) [Fig. 4] performs the adjustment, and a feature extraction section (12) [Fig. ], a word recognition unit (13) that performs word recognition of input speech based on the feature parameters obtained from the extraction unit (12) [Fig. It consists of a candidate creation part (15) that creates candidates for recognized word character strings or recognized syllable characters based on the recognition results from the recognition parts (13) and (14).

Furthermore, in Figure 2, (2) is the main body (1) as shown in Figure 1.
00) is a mechanically and electrically detachable recording/reproducing device such as a tape recorder, 3) is a headphone type microphone as shown in FIG. 1, and (4> is a recording/reproducing device (2) and a microphone (3). ) and the voice recognition section (1) (see Fig. 8). (6) is a display device for displaying character strings etc. generated based on the recognition results; (7) is a keyboard for inputting various control signals of the desk cheating machine;
(8) is a storage device such as a magnetic disk device that stores the character strings generated by the digital cheating machine, and (9) is a speaker (10) that uses the character strings in the storage device by rule synthesis.
This is a speech synthesis unit that reads out the text.

Note that (5) is a control section consisting of a microprocessor, which controls the operations of the above-mentioned sections.

There are two ways to create sentences using the desk cheating machine configured as described above, and each will be explained in detail below.

The first method is to capture the main voice from the microphone (3) into the voice recognition unit (
1), performs voice recognition, converts the input voice into a character string, displays it on the display device (6), and at the same time inputs it into the storage device (8).
).

The second method is to record the text you want to input in advance using a recording and playback device (2
), connect this recording and playback device (2) to this device, and input the recorded text to the voice recognition unit (1) to perform voice recognition and convert the input voice into a character string,
The results are displayed on the display device (6) and simultaneously stored in the storage device (8).

As mentioned above, there are two ways to input audio.
Input switching section (4) In addition to switching the input, the input switching section (4) also controls whether to record the recording signal (a) to the recording/playback device (2) or to the microphone ( 3) Also switches whether to record the input audio.

The operations from voice recording to text creation will be explained in detail below.

(i) Voice registration processing Before performing voice recognition, voice registration is performed in order to create a standard pattern of voice necessary for voice recognition.

First, the syllable registration mode will be described.

The standard pattern described here is the speech recognition unit (1)
It serves as a reference pattern during pattern matching in the phrase recognition unit (14) of the syllable standard pattern memory (14d
) can be stored.

The method of registering voice to this digital cheating machine is to operate the switch (14sl) shown in Fig. 7 and connect the parameter buffer (14d) and the syllable recognition unit (14b).
There are three methods described below.

The first method is to attach a microphone (3) to the main body (100) of the machine.
A method of directly inputting a registered voice, analyzing this registered voice with 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 input the registered voice by connecting the recording and playback device (2) on which the registered voice has been recorded in advance to the main body (100) and playing back the recorded registered voice. This is a method in which the speech recognition unit (1) analyzes the speech recognition unit (1), creates a standard pattern, and stores the created standard pattern in the syllable standard pattern memory (14d) and storage device (8).

The third method is to connect the microphone (3) to the main body (100) of this machine.
At this time, the recording and playback bag e (2) is connected to the main body (100), and while recording this input voice to the recording and playback bag R (2), the registered voice is input directly from the main body (100). ) side analyzes the registered voice from the microphone (3), creates a standard pattern, and stores the created standard pattern in the storage device f! (8). Then, when the audio input to this microphone (3) is completed, the recorded audio is subsequently played back by the recording/playback device (2),
The voice recognition unit (1) analyzes this registered voice,
At the same time as creating a standard pattern and storing the created standard pattern in the syllable standard pattern memory (14d),
This is a method of storing the sound tfJ standard pattern of the directly registered voice from the above-mentioned microphone (3) also in the storage device (8).

In this third method, since the sound recorded on the recording/playback device t(2) is affected by the wave number characteristics between the recording/playback device (2), the standard pattern created from the recorded sound and the microphone ( When comparing the standard pattern created from the voice input directly from step 3), a difference appears between the two standard patterns. Therefore, when recognizing a recorded voice, it is necessary to use a standard pattern created from the recorded voice, and a microphone (3
) to recognize the voice input directly from the microphone (
Since it is necessary to use the standard pattern created from the voice input directly from the microphone (3), by using the method described above, both the standard pattern created from the voice recorded directly from the microphone (3) and the standard pattern created from the recorded voice can be used. Patterns can be created and stored with a single voice registration operation. Also, −
If you record the registered voice on the degree recording and playback device (2), you will not need to input the registrant's voice even on the digital cheating machine that has not created the standard pattern, just by playing back the recorded voice. Standard patterns can be created. Also,
If you record the registered voice on the recording and playback device (2) and also record the text after this registered voice, you can later connect this recording and playback device (2) to the main unit (100) and listen to the recorded voice. Just by playing , everything from voice registration to text creation can be done automatically.

Note that the registrant's voice input for creating the standard voice pattern is performed by the registrant reading out the characters that the present device displays on the display device (6) in a fixed order.

In addition, a recording and playback device (2
), this recording/playback device (2) can be carried alone, by uttering the voice corresponding to the entry word displayed on the display screen and recording it on the recording/playback device (2), the standard pattern can be reproduced. It becomes possible to create.

As mentioned above, when recording the registered voice for creating a standard pattern into the recording/playback device (2), when creating the standard pattern from the recorded registered voice, the recorded voice may be affected by noise etc. There is a possibility that the corresponding headword may be misaligned, and for the sake of explanation, a case will be described below based on FIG. 9 in which a tape recorder is used as the recording/reproducing device. Figure 9(a) shows the state in which registered voices for standard pattern creation are recorded on a tape recorder.
Registered voice “a” ~ “ that corresponds to the headword “a~” “ka”
It shows the state of the tape between "4e" and "o", and here it shows the case where [noise] is completely recorded between "4e" and "o".
When voice registration is performed using a tape with [noise] recorded between the registered voices as shown in Figure 9 (a), the first recorded sound is "a" and the second recorded sound is If you decide which syllable the registered voice that has been input corresponds to simply by the order of the sounds recorded on the tape, such as when a broken sound is "i", it will rise to [noise]. Since it is regarded as voice and the headword is associated with the headword, the actual registered voice input and the headword are misaligned.

Here, FIG. 9(b) incorrectly recognizes [noise] as voice,
This is a diagram in which "Noiseko" is input to the entry word 1e, and the syllable "e" is input to the entry word "o".

In this way, when creating a standard pattern from registered voices, the recorded voice and headwords may deviate due to the influence of noise, etc., so the type of registered voice is changed as shown in Figure 9 (C). Record the indicated character code sound on the recording/playback device (2) in correspondence with the registered voice. With this method, even if Noiseco is recorded between 'u' and 'e', This prevents a discrepancy between the input sound and the headword.

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

First, with reference to FIG. 10, a case will be described in which a recording/playback device with multi-track is used as the recording method.

When a multi-track recording/playback device is used as a recording method, a character hood corresponding to a headword is recorded on a track on which audio is not recorded, as shown in FIG. 2(a). In the voice recognition unit (1), this character
Input from the chord sound.

In addition to knowing the headwords of the sounds that are heard, of the sounds recorded on the audio track, only those that satisfy the condition of sound pressure threshold or higher are recorded in the section t1 in which this kill rafter hood sound was recorded. It is treated as voice and analyzed.

Alternatively, as shown in FIG. 2(b), the character-code corresponding to the entry word is recorded at the beginning and end of the audio, and among the sounds recorded on the audio track, the character-code that indicates the beginning of the audio is recorded. and the back recorded in the interval t2 between the character and chord sound indicating the end of the voice,
Only those that satisfy the condition of being equal to or higher than the sound pressure threshold are regarded as voices and analyzed.

Alternatively, as shown in FIG. 5C, a character food corresponding to the headword is recorded at the beginning of the audio. Among the sounds recorded on the audio track, among the sounds recorded in the section t3 from the character-code sound indicating the type of sound to the character-code sound corresponding to the next headword,
Only those that satisfy the condition of being equal to or higher than the sound pressure threshold are regarded as voices and analyzed.

Also a second method is a single track recording and playback bag!
In case (2), the character code corresponding to the headword is expressed as a sound outside the voice analysis frequency band, and is recorded together with the voice on the track where the voice is recorded. The method for recording character chord sounds in this case is the same as in the multi-track case described above. In other words, the above tl, t
2. Among the sounds recorded in the interval t3, only those that satisfy the same conditions as described above are regarded as voices and analyzed.

However, except in the case of the example shown in FIG.
It is not necessary to use a sound outside the voice analysis frequency band as a chord sound.

Next, alphabets, numbers, parentheses, punctuation marks, etc. are preliminarily stored in the word dictionary (13) of the word recognition unit (13) as shown in Figure 6.
The word standard pattern corresponding to the word registered as a character in d) is stored in the word standard pattern memory (13) in the figure.
c) Register.

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

Next, alphabets, numbers, parentheses, punctuation marks, etc. are displayed on the display device (6) of the device main body (100), and the operator inputs the corresponding pronunciation by voice.

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

Voice recognition becomes possible through the operations described above.

However, when you want to recognize a word that is not in the independent word/adjunct word dictionary (14a) or the word dictionary <13d),
It is necessary to register a word to be recognized in the ancillary word dictionary (14e), or to register an m-word to be recognized in the word dictionary (13d), and a word standard pattern in the word standard pattern memory (13c). However, independent word/attached word dictionary (14
Whether to register the word in c) or to register the word and word standard pattern in the word dictionary (13d) and word standard pattern memory (13c) depends on whether the user wants the word to be recognized as a clause utterance, the word It depends on whether you want to recognize it as a vocalization.

Also, although it is in the independent word/adjunct word dictionary (14e), II
If the word is not in the L word dictionary (13d) and you still want to recognize it by word recognition, it is necessary to register the word and word standard pattern in the word dictionary (13d) and word standard pattern memory (taC).

The method for registering arbitrary words will be described below.

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

When registering a word in the independent word/adjunct word dictionary (14a), the word to be registered is uttered and input into the device.

At this time, this device recognizes this voice with the voice recognition section (1) and displays the recognition result on the display device 1 (6). If the result is correct, the user presses a predetermined key on the keyboard (7) and registers the uttered voice as a character string displayed on the display device (6) in the independent word/adjunct word dictionary (14e). If the recognition result displayed on the display device t(6) is incorrect, the recognition result displayed on the display device (6) is corrected using the syllable correction function of the present device, or the word to be registered is re-uttered. If the re-uttered result is incorrect, it is corrected again using the syllable correction function of this device. The above operations are repeated until the character string displayed on the display device (6) matches the word to be registered.

When registering a word in the FFI language standard pattern memory 3c) and the word dictionary (13d), first register it on the display device (6) in the same way as when registering the word in the independent word / attached word dictionary; 1 (14e). Display the character string you want to register correctly.

Next, the correctly recognized character strings and word standard patterns are stored in the word dictionary (13d) and word standard pattern memory (13
c) respectively.

In addition, recognizing natural voice input is
Considering the current level of speech recognition technology, this is unreasonable. Since the current level of speech recognition technology is limited to continuous syllable utterance input, an example of continuous syllable utterance input will be described below.

In the case of continuous syllable vocalization input, the above procedure is the same, but in the case of continuous syllable vocalization input, the word standard pattern is also a continuous syllable vocalization pattern, so the word you want to register can be re-uttered naturally. , a word standard pattern is created by natural utterance, and the word standard pattern and character string are registered in a word standard pattern memory (13C) and a word dictionary (13d), respectively.

With the above operations, the data necessary for creating sentences using voice recognition has been registered.

(i) Text creation An example of text creation will be described below.

First, when performing a recognition operation, the switch (13sL) of the word recognition unit (13) is set to the parameter buffer (13g).
) and the word determination section (13b), and the switch (14sl) of the phrase recognition section (14) is set so as to connect the parameter buffer (14a) and the syllable recognition section (14b).

There are three ways to write sentences.

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

The second method is to use a recording and playback bag ft (2
) is connected to this device and the recorded text is played back and recognized.

First, an example of online C recognition will be described.

In the case of online recognition, sentences uttered in units of phrases or words are directly input into this device from the microphone (3), so by performing a predetermined operation, the microphone (3) is connected to the voice recognition unit by the input switching unit (4). Connect (1).

Also, if you want to record the audio input from the microphone (3) on the recording/playback device (2),
) to the main unit, and connect the input switching section (4) to the microphone (3).
) and the recording terminal of the recording/playback device (2).

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

During speech recognition, a word recognition unit (13) and a phrase recognition unit (14) are used.
) is running.

The audio input from the microphone (3) is processed by the preprocessing unit (11).
The input audio is processed to have characteristics suitable for audio analysis (for example, when the sound pressure of the input audio is low, the sound pressure is amplified using an amplifier), and the feature extraction unit (
12).

In the feature extraction unit (12), as shown in FIG.
).

At the same time, the analysis unit determination unit (12b) of the feature extraction unit (12)
), based on the analysis results of the analysis unit (12a), the silent period after the utterance of each syllable or phrase, and the beep sound recorded after the utterance of each phrase or word (details will be explained later in offline recognition). ) shown in the embodiment, and when a silent section is detected, a silent section detection signal (b) is generated.

The parameter buffer (12c) that receives the silent section detection signal (mouth) sends the stored feature parameters to the word recognition section (13) and 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. The word determination unit (13b) compares the feature parameters stored in the parameter buffer (13a) with the word standard pattern memory (13c), and selects the feature parameters stored in the parameter buffer (13a) and words with a large likelihood. Words with standard patterns are added to the word dictionary (13d)
A plurality of words are selected, and the character strings of the selected words and their likelihood values are sent to the candidate creation section (15).

On the other hand, the feature parameters that have been input to the clause recognition unit (14) are stored in the parameter buffer (14a). Syllable i! ! In the identification section (14b), the parameter buffer (14
Compare the characteristic parameters stored in the syllable standard pattern memory (14d) with the characteristic parameters stored in the parameter buffer (14d).
The feature parameters stored in m) are converted into a syllable string, and the syllable string is sent to the bunsetsu judgment section (14c).The 0 bunsetsu judgment section (L4c) compares the input syllable string with the independent word/adjunct word dictionary (14e). Compare the words registered in , combine independent words and F1 words to create multiple sets of clauses with a high likelihood, and use the candidate creation unit (1
5) Send to.

The candidate creation unit (15) selects a plurality of characters with a high likelihood from the input character strings, and uses the likelihood value and the data sent from the word recognition unit (13) or the phrase recognition unit 14). A code indicating the data is added and stored. at the same time,
A signal is sent to the control unit (5) to display the character string with the greatest likelihood on the display device. Upon receiving this signal, the control unit (5) inserts a delimiter mark "ma" after the character string with the greatest likelihood.

, and the pair of input sentences shown in FIG. 14(a), for example 1F, are displayed on the display device in the format shown in FIG. 14(C). At the same time, the candidate creation section (15) sends a signal to the control section (5) to cause the storage device (8) to store the contents stored in the candidate selection section (15). The control unit (5) receives this signal and stores the character string stored in the candidate generation unit (15) in the storage device (8) in a form in which a hood representing a delimiter 9 symbol is added after the character string. The character string stored in this external storage device is assumed to be the primary manuscript of the word processor. Generally, a floppy disk is used, but in this case the file format of the storage device (8) must match the file format of the word processor.

In addition, the signal generating section (42) of the input switching section 4) shown in FIG. Enter in (41). The switch (41) connects a circuit so that the audio input from the microphone (3) and the beep sound input from the signal generator (42) are recorded on the recording/playback device e(2), and A beep sound is recorded in a silent section considered to be a break between phrases or words of a sentence recorded in the device (2).

Next, an example of offline recognition will be described.

For offline recognition, this device is equipped with a recording/playback device (2).
Since the text is created by inputting and reproducing the recorded voice, the text is first recorded on the recording/playback device (2).

In addition, in order to input audio from the recording and playback device (2),
The input switching section (4) connects the recording/reproducing device (2) and the speech recognition section (1).

When recording sentences, the system vocalizes phrases or words, creating silent intervals between phrases and words. In addition, when using a dedicated recording/playback device (2) for this device as shown in Figure 1,
In order to clearly mark the divisions between phrases and words, beep sounds indicating divisions can be played on a recording/playback device (2) or with this digital camera.
Separator key (71) set on the machine
) to record.

In addition, the registered words are uttered word by word, but if the recording/playback device (2) has a character-sound generation function and the character-sound corresponding to the Ij and word that you want to input,
If you have one, you can record the character's sound instead of the voice.

In addition, signals indicating the beginning and end of a sentence are recorded together with the voice in order to prevent noise recorded between sentences at the beginning of each sentence from being mistakenly recognized as voice.

However, the recording method of this signal changes depending on whether the recording/playback device (2) is a multi-track system or not. As mentioned in the audio registration section, Fig. 11 shows a multi-track system, and Fig. 12 shows a single-track system. 11(a) and 12(a) show a method of detecting a section in which a sound such as a DTMF signal is recorded as an audio region.

Figures 11(b) and 12(a) show that a sound such as a DTMF signal is recorded before a sentence begins, and then recorded again when the sentence ends, and the section between these two signals is recorded. This is a method of detecting it as a voice area.

Furthermore, in the case of the single track system shown in FIG. 12, a DTMF signal or the like outside the audio band is used, considering that the audio section and the sound such as the DTMF signal overlap.

Furthermore, when recognizing a sentence, the sections t4 and t5 before and after the recorded signal are sampled, and in order to determine whether or not it is voice, it is not always necessary to identify the beginning of the sentence and the beginning of the signal, and the end of the sentence and the end of the signal. They do not need to match, so recognition is possible even if there is a slight lag between the timing at which the sentence is uttered and the timing at which the key is pressed.

Next, the recording/playback device (2) is connected to the main body of this device and the recorded voice is played back to perform recognition processing, but before recognizing this recorded voice, the recognition speed mode is set to a high playback speed of the recorded voice. Then, if you have time and need a high recognition rate, you can select either a fast recognition mode that shortens the recognition time, a mode that performs recognition at normal playback speed, or a double playback recognition mode if you have plenty of time and require a high recognition rate. Set it to mode.

First, an example of the fast listening recognition mode will be described.

In fast listening recognition mode, the playback speed of the recorded audio is increased, so the characteristics of the input audio are different from the standard pattern created from the registered audio played back at the normal playback speed. Even if you input speeded up audio,
Accurate audio! ! can't practice knowledge.

Therefore, in order to accurately recognize audio that has been played back at a faster speed, the sampling frequency is changed. Examples of such methods are described below.

The sampling frequency control unit (12d) of the feature extraction unit (12) in FIG. speed) to sample and analyze the audio. Feature extraction part 〈12
) The subsequent processing is the same as in the embodiment for online recognition. However, a beep sound indicating a break to clarify the break between phrases and words is manually added to the recorded text of the recording/playback device (2), and the feature extraction unit (12) detects the beep sound. At this time, the feature extractor (12) uses the beep sound detection signal (') instead of the silent section detection signal (').
occurs. When the received signal is not a silent section detection signal (b) but a beep sound detection signal (mouth'), the signal generation section (42) of the input switching section (4) indicates a break between clauses or @ words in a sentence. No beep will be generated.

Further, when the speech 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 example of the twice playback recognition mode will be described.

In this mode, the recorded audio is first played back and input to the device.

At this time, the preprocessing section (11) of the speech recognition section (1) reads all the sound pressure fluctuations of the recorded voice and stores this data in the sound pressure fluctuation memory (llb) shown in FIG. Next, the recorded audio is played back and input into the device. At this time, the preprocessing unit (11) uses the data stored in the sound pressure fluctuation memory (llb>) to determine the input sound pressure to the feature extraction unit (12) as shown in FIG. Adjust the amplification factor of the AGC circuit (lla) to 01 to match the appropriate level. That is, the gain G is fixed to the control voltage V. (variable it).

Further, as another example of the twice playback recognition mode, a multiple playback recognition mode can also be considered. This involves playing and inputting a recorded sentence many times, changing the recognition method in the speech recognition unit (1) each time, comparing the recognition results, and selecting the one with the greatest likelihood of certainty. This is the way to do it.

Also, a recording and playback device! (2) When the registration audio is not recorded, and depending on the recording/playback device (2), the frequency characteristics when the playback speed is increased are different from those when the playback speed is normal, Or, when a recorded sentence is recognized by a recording/playback device (2) that has different frequency characteristics from the recording/playback device (2) used to create the standard voice pattern, or when the recording/playback device (2) used to create the standard voice pattern is recognized. ) has the same frequency characteristics according to the standard, but due to the influence of errors in the parts used, the actual frequency characteristics are different from the recording and playback device (2) used to create the standard audio pattern (2) When recognizing recorded sentences, use the function that corrects the effects of frequency characteristics described below.

First, a reference sine wave signal, which is a reference when measuring the frequency characteristics of the recording/playback device (2), is sent to the reference signal generator (42).
is generated and recorded on the recording/playback device fl(2). After that, the recorded reference sine wave signal is reproduced and input to the present device. The input reference sine wave signal is sent to the speech recognition unit (1).
analyzes and determines 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). Correction is applied so as to reduce the difference in frequency characteristics from the reference sine wave signal generated in the section (42). There are many ways to apply the correction, depending on the feature extraction method of the feature extraction unit (12) of the speech recognition unit (1).For example, as shown in FIG.
If it is an analog filter bank system consisting of a parallel connection of an amplifier (AMP) and an amplifier (AMP), the reference signal generator (42) can generate the signal by adjusting the amplification factor of the amplifier (CAMP). The output from the filter is yJ4 so as to reduce the difference in frequency characteristics with the reference sine wave signal.
a. In addition, 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. 12
Various methods can be considered corresponding to the feature extraction method of >.

Through the operations described above, the text input by voice has been converted into a kana string. A correction method when the kana string-converted text differs from the input text will be described below, using FIG. 14 and classifying each case into error. Correct it using the following steps.

Figure 14 (a) is the input text, Figure 14 (b) is the input voice, Figure 14 (c) is the recognition result, Figure 14 (d) to (h) are the correction process,
Figure (i) shows the correction result.

First, we will discuss a correction method when a word uttered is incorrectly recognized as a phrase. As shown in Figure (C), the word "C" uttered is elf! as the phrase "shi-"! If the error occurs, first move the cursor (X) to the part of the incorrect word [(d) in the same figure, then press the word next candidate key (72) to display the next candidate for the word [(d) in the same figure] )il, If this result is correct, proceed to the next part to be corrected. If this result is incorrect, press the word next candidate key (72) again to display the next word candidate. Repeat this operation until the correct answer is displayed.

Next, we will discuss the correction method when a phrase uttered as a phrase ℃ is recognized as a word y4.If a phrase uttered as a phrase 0 is recognized as a word m E I+, first caaf, +c(X ) to the part of the incorrect clause.Next, press the clause next candidate key (73) to display the next clause candidate.If the result is correct, proceed to the next corrected section.

If this result is incorrect, the phrase next candidate key (73)
Press to display the next phrase option. Repeat this operation until the correct answer is displayed.

, #i The previous candidate for each word can be displayed by pressing the pre-word candidate key (74), and the previous candidate for each phrase can be displayed by pressing the pre-phrase candidate key (75).

If the correct answer cannot be obtained using the above two correction methods, correction may be performed in units of syllables, or words, phrases, or syllables may be re-inputted.

In addition, when re-voicing is input, the phrase can be word-recognized again, and I
In order to avoid recognizing K words as phrases, the candidate generation section (15) considers only the recognition results sent from the word recognition section (13) as the recognition results, and recognizes the phrase recognition section (14).
) can be externally controlled to be ignored.

In addition, the candidate generation unit (15) is configured to consider only the recognition results sent from the phrase B placement unit (14) as recognition results, and to ignore the recognition results sent from the word recognition unit (13). More control.

The above-mentioned next candidate key is a key having the function described below, and will be explained using FIG. 15.

In the speech recognition unit (1) of this device, 11 word recognition and bunsetsu recognition run in parallel, and as mentioned above, the re-recognition results for words and clauses are obtained. The phrase next candidate key (73) is used to display the recognition results on the display device (6) in order of the recognition results, starting with the words with the highest likelihood. The word next candidate key (72) is used to display recognition results on the display device in order from ) are the pre-word candidate key and the pre-clause candidate key.

FIG. 15 shows the candidate buffer (15a) of the candidate creation section (15).
). In this figure, the negative recognition result is "tango", and this is represented by (word) that this is the recognition result sent from the word recognition unit (13). Similarly, the recognition result of the third place is 'Tango wo', and this is the recognition result sent from the sentence m recognition unit (No. 1), which is expressed as (bunsetsu), and the recognition result of the third place is , “Tangoni, desu,
This is the recognition result sent from the phrase recognition unit (14), which is expressed by (clause), and the fourth recognition result is ``Tankou.''

This is the recognition result sent from the word recognition unit (13), which is represented by (word).

Display now! It is assumed that "tango" is displayed in (6). In this state, when the phrase next candidate key (73) is pressed, 'Tango wo' is displayed on the display device (6). Further, when the next word candidate key (72) is pressed, "tankou" is displayed on the display device (6).

In addition, when 'Tanko' is displayed on the display device (6), if you press the word front candidate key (74), 'Tango' will be displayed on the display device (6), and the sentence front candidate will be displayed. key (7
3) When the button is pressed, "Tangoni" is displayed on the display device (6).

Next, we will discuss how to modify an entire passage at once.

The example in FIG. 14(e) is an example in which the word ``1T'' is incorrectly recognized as r A . First, move the cursor to the word you want to correct [Figure 1 (e) 1.

Next, press the word next candidate key (72) to display the next word candidate [FIG. 4(e) il]. If this result is correct, proceed to the next modification part. If this result is incorrect, the next word candidate key (72) is pressed to display the next word candidate. Repeat this operation until the correct answer is displayed. If the correct answer is not displayed, re-speak and re-enter. By pressing the previous word candidate key (74), candidates for the word displayed immediately before can be displayed.

Next, we will discuss how to correct an entire word at once.

The example in FIG. 14(r) is an example in which the phrase "gamen no" is incorrectly recognized as 1 gamen no. First, move the cursor to the phrase you want to modify [FIG. 1(f) il.

Next, press the phrase next candidate key (73) to display the next phrase candidate. If this result is correct, proceed to the next modification section. If this result is incorrect, press the phrase next candidate key (73) to display the next phrase candidate.

Repeat this operation until the correct answer is displayed. If the correct answer is not displayed, re-speak and re-enter. Press the previous syllable candidate key (75) to select the previous syllable candidate. It can also be displayed.

Next, a method for correcting the u-joint position will be described.

The example in Figure 14 (h) is an example in which the phrase ``Onseide'' is incorrectly recognized as ``Onkeide.'' In this example, the syllable ``ke'' is corrected to ``1 se''. Move to the 1st syllable you want to correct (X) and press the next syllable candidate key (76). By pressing the syllable next candidate key (76), the syllable closest to the syllable of the part you want to correct will be displayed.
Move to the next modified part. If this result is incorrect, press the syllable next candidate key again to display the next syllable candidate.

Repeat this operation until the correct answer is displayed. If the correct answer is not displayed, re-enter by speaking again. If the re-input results are incorrect, correct them again using the steps above. Repeat this operation until the correct answer is displayed.

It is also possible to avoid displaying the previous syllable candidate by pressing the previous syllable candidate key (77).

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

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

Next, using FIG. 16, the several syllable correction method will be described.

In this example, the input sentence "Kaijiyo" shown in FIG. In this case, first move the cursor (X) to the syllable you want to correct [
(C) of the same figure], re-input ``Kai゛°''.The re-uttered input voice is recognized by the speech recognition unit (1), and the recognition result is displayed on the display device (6).If the recognition result is correct, If you misrecognize ``kai'' as ``kae'' as shown in Figure (d), if it is a word, press the word next candidate key (72>. In this case, press the phrase next candidate key (73). Since Figure 16 is an example of a word, the method for modifying a word will be described below.
If you press the next word candidate key (72) in the state of d),
First, the control unit (5) selects from the word dictionary (13d) the recognition result of the figure (b) before correction, ``1 Ganjo'', and the recognition result of the figure (d) after re-voicing, ``Kaejiyo''. Compare the numbers and find the molecular size of the corresponding part. Next, the control unit (5) selects a word having the same part "1" from the word dictionary (13d). Figure (f') shows the stored contents of the word dictionary (13d), and Figure (g) shows words with "jiyo" selected from the stored contents. Next, the control section (5)
is the word written in (g) of the same figure and the recognition result after re-uttering "
Calculate the likelihood of ``Kaejiyo'' and display the word with the largest likelihood value [(e) in the same figure.

Next, we will discuss the case of correcting recognition boundary errors of phrases or words.

In the example in Section 14 (g), the phrase ``Bunsyowo'' is incorrectly recognized as having a silent interval between ``n'' and ``shi'', which is marked with [].
This is an example where the word ``fun'' and the phrase ``you'' were separated into two parts and were misrecognized.

In this case, the recognition boundary error must be corrected, but if you want to delete the recognition boundary delimiter, move the cursor (X) to the recognition boundary delimiter you want to delete [see figure (g)
il, press the delete key (78) [(g) il in the same figure. If you want to insert a recognition boundary delimiter, move the cursor (X) to the syllable at the position you want to insert and press the insert key (79).

However, as will be described later, the delimiter beep of the recording/playback device (2) and the delimiter added to the recognition result stored in the storage device (8) are different from each other. This will serve as a landmark for synchronization, so you must take precautions.
) indicates that a delimiter has been inserted or deleted in the storage device (8).
Remember it.

For example, the sentence shown in Figure 14(g)i is
g) Assume that the information is stored in the storage device (8) as shown in 1. When the sentence in (g)i is modified as shown in (g)i, the delimiter rv, which was completely stored in the storage device (8), is changed to the symbol "ma". The symbol "1" is a symbol that indicates that the delimiter 1'y" has been deleted, and is not used as a symbol to indicate a recognition unit, but is a symbol used only for control with the recording/playback device (2), etc. be.

With this configuration, even after deleting the delimiter 19, the beep sound produced by the O sound reproduction device (2) and the symbol mV that has been completely stored in the storage device (8).

'1. By using , both devices can be controlled in synchronization.

The above is an example where the delimiter "wa" is deleted, but the same idea can be applied when it is inserted. In other words, a specific symbol that is not used as a control signal and only represents a delimiter is called the delimiter "ma".

You can insert it instead of .

By the above correction procedure, as shown in FIG. 14(i),
Correct the text.

After the recognition boundary error has been corrected, the recognition unit for which the recognition boundary error has been corrected is corrected according to the correction procedure. In the case of correction by re-voicing, the voice of anyone who has registered the standard pattern can be recognized, so corrections can be made even if one is not the person who recorded the text.

The method for correcting kana string sentences has been described above, and the following functions are provided to assist in correction.

By moving the cursor on the character string displayed on the display device (6) and scrolling the display screen, it is possible to sequentially display memorized sentences from the storage device (8) on the display screen, but at this time, the memorized sentences can be displayed on the display screen in sequence. The audio corresponding to the part is played back from the recording/playback device (2).

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

In addition, in both of the above methods, the time before the delimiter recorded in the recorded text and the delimiter recorded on the display side are used as timing signals for synchronization, and the playback of the recording and playback device (2) is performed. and the display are controlled so that they operate in synchronization with each other. Also, the keyboard (7
), or when a signal to stop the playback is input from the recording/playback device (2), the playback is stopped and the scrolling of the display or the movement of the cursor is also stopped.

The above synchronization function between playback and display of the recording/playback device (2) allows character strings to be checked while listening to the playback sound, making it easier to find corrections.

As for the synchronization method described here, the character string of the storage device (8>) corresponding to the part that has been completely played is transferred to the display device (6
), and another method is to display on the display device a kana sequence of a part that is one delimiter later than the part corresponding to the part that is being played back.

In this case, when the display is stopped for correction, the corrected part of the recorded voice has already been played back, so in order to play the corrected part again, it is necessary to locate the beginning of the part to be corrected from the reproduced text. Therefore, when using this method, when the display is stopped, the recording/playback device (
1 (2) to provide a function to backtrack to the previous delimiter.

In addition, when a tape recorder is used as the recording/playback device (2), the playback section is controlled by the rotation of a motor, and the beginning of the section corresponding to the correction section may not be accurately located due to slack in the tape, etc. There is.

In such a case, record the input audio as PCM recording or ADPCM recording for a certain period of time, and if you want to listen back to the input audio, listen back to the PCMfl sound or ADPCM recorded audio. Add functionality.

1. Fig. 17 is an example of the above-mentioned function, and PCM0
Numbers 01 to 05 in Figure 0, which is a diagram of a PCM data memory that stores sound data, indicate addresses. The input voice is “Washiwa1ten1shi-1” shown in Figure 14.
The sentence is 1 marupa, which is 1 bunsyo of aru 1 te-1 kamen.

When the above audio is input, the PCM data memory (
DM), the voice "Washiwa" up to the first silent section is stored at address 01. At address 02, the sound "ten" up to the second silent section is stored. At address 05, the sound "te-°" up to the fifth silent section is stored. At this time,
The PCM address pointer (AP) stores the address of the first data stored in the PCM data memory. In this example, 01 is stored.

At this stage, the PCM data memory is full.

Next, when audio is input, the PCM data memory (
The input voice is stored at the address of the data stored first among the data stored in DM>. In this example, “Gamen no” is stored at address 01 where “Washiwa °” was stored. At this time, the PCM address pointer (AP) is set to the PCM data memory (DM).
In this example, 02 is stored as the address of the data stored first among the data stored in .

In this state, when reproducing the contents of the PCM data memory (DM), the contents are reproduced starting from the address pointed to by the PCM address pointer (AP). In this example, 02, 03,
It will be played back in the order of 04, 05, and 01.

With this method, it is possible to listen back to the audio accurately and quickly as many times as desired.

Also, move the cursor (X) to the recognition unit delimiter on the screen.
) and press the start key (70) of the recorded audio to search the recorded text for the part behind the delimiter on the recording/playback device (2) that corresponds to the recognition unit indicated by the cursor, and then press this key. It has a function to play the following sentences. An example of such functionality is shown below.

In order to confirm the recognized sentences, record the recognition results on a device 8.
) and displayed on the display device (6) from the beginning. At this time, as shown in FIG. 19, the delimiter counter (5a) of the control unit (5) counts the number of delimiters that have been read out from the storage device (8).If the read recognition result is incorrect, places the cursor on the incorrect part and presses the cue key.The control unit (5) causes the recording/playback device (2) to play back the recorded text in fast-forward playback mode. The beep sound counter (12e) of the feature extraction unit (12)
) counts the beep sounds without marking the breaks in the sentence input from the recording/playback device (2).

The comparator circuit (5b) sends the signal (c) to the recording/reproducing device (2) when the value of the beep counter (12e) becomes one less than the value of the delimiter counter (5a) mentioned above. send and stop playback.

In addition, in order to confirm the recognition result and the corrected text, the data stored in the storage device (8) can be displayed on the display device (6).
It is very tiring for the eyes because the user has to visually follow and read the string displayed on the display screen.

In view of this, this device is a memory device that stores recognition results! (8) By providing a function to read out the above character string using a voice synthesis function, it is possible to check the recognition result and the corrected text by listening to the voice synthesized sound.

In this case as well, the delimiter is used as a timing signal for synchronizing the speech synthesis section (9), the storage device (8), the recording and reproducing device (2), and the display device e (6).

In other words, the character string corresponding to the part that the speech synthesis unit (9) is reading out from the storage device (8) is displayed on the display device (6), and at the same time, the recording part is cued up by the recording and playback device (2). . With this method, when an error is discovered by the voice synthesis function and the voice synthesis function is stopped to correct it, both the display on the display device (6) and the recorded part on the recording and playback device m(2) are incorrect. It shows the part,
Corrections can be made instantly.

The method of synchronization described here is to display on the display device (6) the kana column of the storage device corresponding to the part being read aloud by the speech synthesis function, and at the same time display the kana column on the display device (6).
The method of reproducing the corresponding syllable part from the recorded sentence in (2), and the part corresponding to the part being read out by the speech synthesis function, are recorded on the recording and playback device (2) one delimiter later than the part corresponding to the part being read out by the speech synthesis function. There is a way to play back the parts of the text. In the latter case, when voice synthesis is stopped for correction, the recording and reproducing device (2) has stopped before the part to be corrected, so if it is played back in this state, the corrected part of the audio can be immediately played back. In the former case, when g-voice synthesis is stopped for correction, the corrected part of the recorded voice has already been played back, so it is necessary to backtrack in order to play the corrected part again. Therefore, when adopting the former method, when the display is stopped, the recording and playback device (2) is automatically activated.
It is preferable to have the function of backtracking to the previous delimiter.

Above, an example has been described in which the recognition result is stored in the storage device (8), but below, as another example, a method of storing the recognition result in the recording/reproducing device (2) will be described.

Memory device f! The recognition results stored in (8) are recorded in the recording/playback device (2) that has finished recording the original text. In this way, the original text and the recognition results can be recorded on the same recording medium, making it easier to manage the original text and the recognition results.

In addition, while recording and inputting, you can record and playback the recognized results! By recording in step (2), an external storage device becomes unnecessary.

In either case, a multi-track recording/playback device (
By using 2), it is possible to store the recognition result in a track on which no audio is recorded while reproducing the recorded audio.

(See the margin below.) Effects of the Invention When a sentence recorded on a tape recorder or the like is input into a speech recognition word processor, the playback speed was conventionally equal to 0 sound speed, so the recognition time was equal to the recording time of the recorded voice. However, according to the speech recognition system of the present invention,
This recognition time can be shortened. In addition, simply inputting the reproduced voice into a recognition device by increasing the playback speed of the recorded voice compared to the speed at which it was recorded will not allow the voice to be recognized with a high recognition rate due to the effect of a shift in the frequency band of the voice. Although it is possible, by changing the sampling frequency, a highly accurate speech pattern can be obtained without deteriorating the recognition rate.

[Brief explanation of the drawing]

Fig. 1 is an external view of a daytime cheating machine that employs the speech recognition system of the present invention, Fig. 2 is a block diagram of the daytime coaching machine, Figure 3 is a block diagram of the voice recognition section (1), and Figure 4 is a block diagram of the daytime coaching machine. The figure is a block diagram of the preprocessing unit (11), Figure 5 is a block diagram of the feature extraction unit (12), Figure 6 is a block diagram of the word recognition unit (13), and Figure 7 is a block diagram of the phrase recognition unit (14). Figure 8 is a diagram showing the configuration of the input switching unit (4), Figure 9 is a diagram showing the relationship between headwords, recording methods, and character totals, and Figure 10 is the relationship between character and sound recording methods and voice sections. Figure 11 is a diagram showing the recording method when the recording/playback device is a multi-track system, Figure 12 is a diagram showing the recording method when the recording/playback system is a single-track system, and Figure 13 is an example of a frequency correction circuit. , FIG. 14 is a diagram showing the correction at the time of misrecognition, FIG. 15 is a diagram showing the candidate buffer (15a) in the candidate creation unit (15), and FIG. 16 is an example of correcting several syllables at the time of misrecognition. ``Figure 1 and Figure 17 are explanatory diagrams of the PCM recording method, Figure 18 is an explanatory diagram of AGC operation, and Figure 1913 is an explanatory diagram of the delimiter counter. (1)...Audio E 2 parts, <z>-6R'8-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, <Ll
a)...Variable gain amplifier, (llb)...Sound pressure variation memory. Fig. 2 Fig. 4 Fig. 11 Preprocessing section Fig. 17 (a) (b) Fig. 6 Fig. 7 Fig. 8 Fig. δ Fig. 10 1 Ten 1 Ten 1 Ye -1 Ah, 1, 1 Kamen 1 Funbunsei, 1, 1 φ, 1 φ Useful 1 Maru 1 Maru 1 Maru I'm V, I'm V, 91iQ? H9AV...i...
C” RvT9 Kamen's ma...i...Gamen's mabunguchiyo 7...(h) i...Bunshotwo, Onboyide 9...i・・・Bunsyo wo ma, ”i ′Bia, 1 is V...ma v
vvv (i) I, CRT Kamen's V function, 7, V, V, who was sleeping on my head, V (j) i...Gamen's 9 times, 7 times.
・・・・i・・・・Gamen no 7 bun 7 yo ・・・
・Illustration of Figure 18

Claims (1)

[Claims] (1) Playback audio obtained from the recording and playback device that is equipped with a recording and playback device with a variable playback speed and a voice recognition device that recognizes the sound that is played back and input from the recording and playback device, and that changes the playback speed with respect to the recording speed. A voice recognition system that recognizes.