JPH0632005B2 - Japanese voice input device - Google Patents

Description

Description: TECHNICAL FIELD OF THE INVENTION The present invention relates to an improvement of a Japanese voice input device for recognizing an input voice in syllable units, and more specifically, inputting a single syllable voice for one syllable. The Japanese speech input device suitable for recognizing continuous speech is provided with the generated characteristic pattern and the characteristic pattern obtained from the uttered continuous speech as characteristic standard patterns.

<Technical Background of the Invention and Problems> In a conventional voice recognition device, a registration mode for registering a voice feature standard pattern and a recognition mode (input mode) for recognizing an input voice are separated and input in the recognition mode. The characteristic pattern obtained by analyzing the generated voice could not be registered as the characteristic standard pattern. This is not a problem when using words as the unit of recognition,
When syllable is used as a unit, each syllable is affected by the preceding and following syllables (articulation coupling), and the difference in the strength and height of the voice depending on the syllable position in the word or syllable is also severe, resulting in deterioration of recognition performance. There was a problem.

In order to solve such a problem, in the past, one syllable was registered several times at the time of registration, and it was dealt with by having many characteristic standard patterns. It was impossible to pre-register feature standard patterns for syllable environments in all cases.

<Objects of the Invention> The present invention has been made in view of the above points, and in a voice input device having a plurality of characteristic standard patterns for one syllable, with respect to a characteristic standard pattern registered in advance. The purpose is to improve the recognition performance by enabling the registration of the characteristic pattern of the desired syllable of the characteristic pattern obtained by analyzing the continuous voice captured at the time of inputting the recognition mode, and in order to achieve this purpose, The Japanese voice input device of the present invention recognizes an input voice and a first memory in which a feature standard pattern of a monosyllabic voice input in a registration mode for registering a voice feature standard pattern is registered. In the mode, a temporary storage memory for temporarily storing the waveform and characteristic pattern of each syllable of the continuous speech of the input phrase, and a display for displaying the recognition result of the input speech. A step, a correction means for correcting the recognition result as necessary, an instruction means for instructing registration of a characteristic pattern of a desired syllable in a continuous speech of an input phrase as a characteristic pattern in the recognition mode, Reproduction output means for reading out the waveform of the temporary storage memory corresponding to the syllable instructed by the instructing means and outputting the sound, and, after reproduction by the reproduction output means, instructing execution of registration of the characteristic pattern of the syllable instructed by the instructing means. And a second memory in which the characteristic standard pattern of the temporary storage memory obtained by analyzing the continuous voice based on the instruction of the execution instructing means is registered as the characteristic standard pattern, After the desired syllable in the phrase input in the recognition mode is displayed and corrected for playback confirmation, this feature pattern can be registered as a feature standard pattern. Those were.

Further, according to the embodiment of the present invention, the syllables of the feature standard pattern that can be registered are limited by using the recognition result of the voice instructed to be registered as the feature standard pattern, and the extremely different voice feature patterns are featured. It is designed so that it does not become a standard pattern, and by further storing the input voice, the voice of the part corresponding to the syllable instructed to be registered as a characteristic standard pattern is reproduced and output, and is accurately cut out as a syllable. It is designed so that the input person himself can confirm whether or not the input has been made.

<Embodiment of the Invention> Hereinafter, with reference to the drawings, the present invention recognizes continuously uttered speech in syllable units, corrects the recognition result with an input device such as a keyboard, and then, in units of words or the like. A Japanese voice input device having a function of transferring to an external device will be described as an example.

FIG. 1 is a block diagram showing the configuration of an apparatus according to an embodiment of the present invention.

In FIG. 1, a voice that has been uttered and input is input to an analog input section 2 via a microphone 1 or the like, amplified by an amplification section 3 in the analog input section 2, and then digitalized by an analog / digital conversion section 4. The signal is converted into a signal, and the digital signal is input to the voice analysis unit 5 and the syllable segmentation unit 6.

The voice analysis unit 5 divides the input voice into frames of about 16 ms, performs spectrum analysis, and at the intervals of about 8 ms, the syllable segmentation unit 6 features patterns and information necessary for syllable segmentation (power, number of zero crossings, etc.).
To transfer.

The syllable segmentation unit 6 cuts out a syllable from the input voice using various information from the voice analysis unit 5. The characteristic pattern of the cut-out portion and the voice waveform of the section are stored in the waveform / characteristic pattern temporary memory 7. Then, the fact that the syllable is cut out is transmitted to the CPU 8 and the address in the waveform / feature pattern temporary memory 7 is transmitted at the same time.

The waveform / feature pattern temporary memory 7 is configured to store a plurality of syllables.

The processing of the syllable segmentation unit 6 is configured so that the start / stop is controlled by an instruction from the CPU 8.

9 is a monosyllabic recognition unit, and the monosyllabic recognition unit 9
A characteristic pattern memory 10a in the pattern memory 10 and standard pattern memories 10b and 10 in response to a command from U8.
The distance calculation and the like are performed with c and the result is returned to the CPU 8. Then, the CPU 8 stores the result in the recognition result storage memory 11 and displays it on the display device 12. The recognition result storage memory 11 is configured to store the recognition results for a plurality of syllables.

Reference numeral 13 is a voice output control unit, and the voice output control unit 13 sends information stored in an arbitrary portion of the waveform / feature pattern temporary memory 7 to the analog output unit 14 in accordance with a command from the CPU 8. The analog output unit 14 converts the audio digital signal into a digital-analog conversion unit 14a and an amplifier 1.
4b to reproduce and output a voice waveform.

The pattern memory 10 is divided into three parts, and 10a is a characteristic pattern memory and can store only one characteristic pattern corresponding to the input syllable. The latter two 10b and 10c are memory for characteristic standard patterns, the area 10b is memory for characteristic patterns of syllables registered in the registration mode, and the area 10c is the recognition (input) mode realized by this patent. This is a memory for the characteristic patterns of syllables registered in.

The syllable characteristic pattern memories 10b and 10c are each composed of an area for storing each syllable name as a code, a flag area for storing the presence or absence of registration, and a characteristic standard pattern area for storing characteristic standard pattern data, as will be described later. There is.

Reference numeral 15 is an input unit composed of a keyboard or the like, and is provided with, for example, a kana key 15a, a registration mode key 15b, a recognition mode key 15c, a syllable registration key 15d, etc., as shown in FIG.

Reference numeral 16 is an I / F unit that controls transmission / reception of data when the recognition result is transferred to an external device.

Next, the operation of the apparatus configured as described above will be described in the registration mode and the recognition mode.

I. Description of Registration Mode FIG. 3 shows a processing flow of the CPU 8 in the registration mode.

In FIG. 3, when the main body of the apparatus is set to the registration mode by operating the registration mode key 15b, first step n1
At, the pattern memory 10 is initialized and all standard patterns are erased. Table 1 shows the standard pattern memory 10
2 shows the configuration of the standard pattern memory 10
c has the same structure.

The initialization process in step n1 is realized by putting "0" in the flag area of presence / absence of registration in the standard pattern memories 10b and 10c. Next, in step n2, the monosyllable to be uttered is displayed on the display device 12 as follows.

"1" of "A _1" subscript here shows that it is a one-th in the pattern of "A".

The operator looks at the display of the display device 12 and utters a predetermined monosyllabic voice to input.

In response to this voice input, the process proceeds to step n3, where the syllable segmentation unit 6 is instructed to start the cutout of voice, and the syllable segmentation unit 6 cuts out a single syllable and is obtained by the waveform and voice analysis unit 5 of the section. The feature pattern is stored in the waveform / feature pattern temporary memory 7.

In step n4, the syllable segmentation unit 6 checks whether or not a single syllable has been cut out, and when cut out, the process proceeds to the next step n5.

In step n5, the syllable segmentation unit 6 is instructed to stop the cutout processing, and the registration processing is continued.

In step n6, the voice portion corresponding to the syllabary just cut out is read from the waveform / feature pattern temporary memory 7 and reproduced and output from the analog output unit 14 via the voice output control unit 13.

In step n7, it is determined whether or not the operator has correctly cut out based on the reproduced and output voice, and whether to perform recutting or registration is determined according to the instruction from the keyboard 15 as a result. In step n7, when the operator hears the reproduction output and determines that the cutout has been performed correctly, the execution key 15i is operated, and the process proceeds to step n8. When the operator instructs recutting, the release is performed. In response to the operation of the key 15h, the process returns to step n3.

At step n8, the characteristic standard pattern is stored at the position of the characteristic standard pattern memory 10b corresponding to the syllable displayed on the display device 12, and "1" is set to the flag indicating the presence or absence of the corresponding registration.

In step n9, it is determined whether or not the registration of all the standard patterns has been completed.
To return to the 2, the display of the next single syllable, for example, to display the "A _2", performs the same processing.

In this way, when registration is completed, several standard feature patterns of all monosyllabic patterns are registered in the standard pattern memory 11b.

Next, the operation of the recognition mode will be described.

II. Description of Recognition Mode FIG. 4 shows a processing flow of the CPU 8 in the recognition mode.

First, when the apparatus is set to the recognition mode by operating the recognition mode key 15c and the operator utters a voice to be recognized, in step n11, in response to this input voice, the syllable segmentation unit 6 is instructed to start syllabic segmentation. Then, the syllable segmentation unit 6
The characteristic pattern temporary memory 7 is initialized, and the characteristic patterns and waveforms corresponding to the syllabaries cut out thereafter are entered from the head address, and the information of the waveform of each syllable and the start and end addresses of the characteristic pattern is given to the CPU 8.

In step n12, it is checked whether or not the syllable is cut out, and when it is cut out, the process proceeds to step n13.

At step n13, the waveform / characteristic pattern temporary memory 7
The characteristic pattern is transferred to the area of the characteristic pattern memory 10a of the pattern memory 10 for recognition. That is, by giving a recognition command to the monosyllabic recognition unit 9, the contents of the characteristic pattern memory 10a and the standard pattern memories 10b, 10
Recognition is performed by collating the contents of c, and the result is stored in the recognition result storage memory 7 and displayed on the display device 12 (step n14).

For example, if the first rank of the recognition result when uttering "Kaimono" as the input voice is "Kagimo", the display device 12
Is displayed as a key, and a plurality of recognition result candidates for each syllable are stored in the recognition result storage memory 11 as shown in Table 2, for example.

When the input of the word such as "Kaimono" is completed, the operator inputs "End" key 15g of the keyboard input unit 15.
Enter. Then, the syllable segmentation unit 6
To stop cutting (steps n15, n1)
6). Then, if all the character strings are correct, by inputting the "transfer" key 15j, kana characters can be output to the external device via the I / F unit 16 (step n1).
8, n19).

If you look at the display of the recognition result and find that most of the characters are incorrect or you make a mistake, you can press the "Cancel" key 15f.
By inputting, by the judgment in step n17,
The initial state can be restored.

If a part of the recognition results is different, the operator will make a correction with the key as shown in step n20.

There are two ways to modify by key.

First, move the cursor to the position you want to correct with the cursor movement keys 15k, 15
l Use "→" and "←" to move the cursor to the character you want to correct. For example, when it is desired to correct the second character "Gi", the display by operating the cursor movement key "←" 15l is as follows.

One method is to use the keyboard 1 at this cursor position.
The following corrections are made by inserting characters with the No. 5 kana key 15a.

Another method is to enter the “syllabic candidate” key 15e on the keyboard 15 Can be modified as follows.

Here, in the case of registering the input voice of "i" as a standard pattern according to the present invention, by inputting the "syllabic registration" key 15d, the process shifts to step n22 in the judgment of step n21, and the voice output control unit Indicate the start and end addresses of the waveform of the syllable to be reproduced in 13
The waveform of the corresponding portion is read from the characteristic pattern temporary memory 7, and the analog output unit 14 is passed through the voice output control unit 13.
And the sound corresponding to “i” is reproduced and output from the analog output unit 14.

By listening to this reproduced voice, the operator determines whether the cutout position is correct by the input person himself,
When the "execute" key 15i or the "cancel" key 15h is input, the process proceeds from step n23 to step n24 when the "execute" key 15i is pressed, and to step n17 when the "cancel" key 15h is pressed. become.

CPU moves from step n23 to step n24
Reference numeral 8 registers the feature pattern of the syllable stored in the waveform / feature pattern temporary memory 7 as the syllable feature pattern corresponding to the designated voice in the standard pattern memory 10c of the pattern memory 10. At this time, syllables other than the syllable candidate group of the recognition result cannot be registered. That is, the feature pattern for each syllable stored in the waveform / feature pattern temporary memory 7 can be registered only as the feature standard pattern for each syllable candidate of the recognition result. Therefore, for example, in the present example, the characteristic pattern of the first syllable cannot be registered as a characteristic standard pattern other than "ka", "ta", "ga", and "ha".

Table 3 shows a configuration example of the standard pattern memory 10c.

In the above example, since the input voice is registered as "i", the feature pattern is transferred to a place where there is no registration in the syllable name "i", that is, "i _c ", and registration is performed as "1".
To

If the presence or absence of registration is all “1”, for example, in the case of “A” in Table 3, up to “A _c ” is registered, so the area of “A _a ” that is the oldest in terms of time Delete the data in and then register at that location. In the above embodiment, the standard pattern memory is divided into 10b and 0c and registered, but the present invention is not limited to this. For example, by judging the quality of the standard pattern, the worst characteristic standard pattern is erased, The syllable at the time of input may be registered in that area.

As a method for judging the quality of the standard pattern, for example, the method previously proposed by the present inventors as Japanese Patent Application No. 57-217296 "Voice recognition device", that is, a counter means is provided for each characteristic standard pattern to recognize the input voice. Depending on the judgment result,
There is a method of increasing or decreasing the counter value and determining the worst characteristic standard pattern according to the counter value.

As described above, according to the above-described embodiment, in the voice input device having a plurality of characteristic standard patterns for one syllable, the characteristic pattern of the voice captured at the time of input is added to or replaced with the characteristic standard pattern. By doing so, the recognition performance can be improved.

Further, by using the recognition result of the voice instructed to be registered as the characteristic standard pattern, the syllables of the characteristic standard pattern that can be registered are limited so that extremely different voice characteristic patterns do not become the characteristic standard pattern. You can This is to check whether or not the syllable is cut out at a proper position when cutting out the syllable from the voice.
For example, if you want to register the third syllable "Momo" when you say "Kaimono" as input, "Gi" is the first recognition result, "Ri" is the second recognition result, and "I" However, according to the embodiment of the present invention, the syllable displayed as a candidate of the recognition result may be inconvenient if the characteristic pattern of the voice is registered as the characteristic standard pattern of “mo”. Since it is possible to register only for
Such inconvenience does not occur.

Further, by storing the input voice, in the case of the above example, when "mo" is instructed to be registered as a characteristic standard pattern, the voice corresponding to the syllable is reproduced and output, and is accurately converted into a syllable. It is possible for the input person to confirm whether or not it has been cut out, and it is possible to avoid registering an erroneous feature standard pattern.

<Effects of the Invention> As described above, according to the present invention, in the Japanese voice input device for recognizing the input voice by the syllable unit by collating with the pre-registered characteristic standard patterns of plural kinds of syllables, In the registration mode for registering the characteristic standard pattern, the first memory in which the characteristic standard pattern of the monosyllabic voice input in the registration mode is registered, and in the recognition mode for recognizing the input speech,
A temporary storage memory for temporarily storing the waveform and characteristic pattern of each syllable of the continuous speech of the input phrase, a display means for displaying the recognition result of the input speech, and a correction means for correcting the recognition result as necessary. An instruction means for instructing to register a characteristic pattern of a desired syllable in a continuous speech of an input phrase as a characteristic pattern in the recognition mode, and a waveform of a temporary storage memory corresponding to the syllable instructed by the instructing means. Based on an instruction of the execution output means, a reproduction output means for reading out and outputting a sound, an execution instruction means for instructing execution of registration of a characteristic pattern of a syllable instructed by the instruction means after reproduction of the reproduction output means, The feature standard pattern of the temporary storage memory obtained by analyzing the continuous voice is provided with the second memory registered as the feature standard pattern. This feature is characterized in that, after the desired syllables in the words and phrases that have been input at the time have been displayed and the reproduction has been confirmed, this feature pattern can be registered as a feature standard pattern. In contrast, you can add or replace the characteristic pattern of the voice captured at the time of input as a standard pattern,
As a result, the recognition performance can be improved.

That is, when using a voice input device, syllables that are difficult to input may appear. For example, when singing "i" can be recognized as "i", when singing "kai" The phenomenon that "i" is always mistakenly recognized as "gi" may occur. This "i" is a second syllable of a word consisting of four syllables, with a syllabic "mo" before and after "ka", and is said to be uttered at a higher voice than "i" singed alone. Although it is in a syllable environment, in the present invention, it becomes possible to put the characteristic pattern of "i" uttered in this syllable environment into the standard pattern, and uttered in the same or similar syllable environment thereafter. Can be accurately recognized.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an apparatus according to an embodiment of the present invention, FIG. 2 is a plan view showing an example of an input unit such as a keyboard, and FIG. 3 is a process flow chart for explaining an operation in a registration mode. , FIG. 4 is a processing flow diagram for explaining the operation in the recognition mode. 5 ... Voice analysis section, 7 ... Waveform / feature pattern temporary memory, 8 ... CPU, 9 ... Single syllable recognition section, 10 ... Pattern memory, 10a ... Feature pattern memory, 10b and 10c ... Standard pattern Memory, 11 ... Recognition result storage memory, 15 ... Keyboard, 15b ... Registration mode key, 15c ... Recognition mode key, 15d ... Syllable registration key.

 ─────────────────────────────────────────────────── ─── Continuation of front page (56) Reference JP-A-58-62699 (JP, A) JP-A-58-62739 (JP, A) JP-A-58-195955 (JP, A) JP-A-58- 220197 (JP, A)

Claims (1)

[Claims] 1. A Japanese voice input device for recognizing an input voice by a syllable unit by collating input voices with a plurality of pre-registered feature standard patterns of syllables, in a registration mode for registering voice feature standard patterns, A first memory in which a standard feature pattern of a single syllable input by a single syllable voice is registered, and a waveform and a characteristic pattern of each syllable of a continuous voice of an input phrase are temporarily stored in a recognition mode in which the input voice is recognized. A temporary storage memory, a display unit for displaying the recognition result of the input voice, a correction unit for correcting the recognition result of the input voice if necessary, and a continuous voice of the phrases input in the recognition mode. Instructing means for instructing registration of a characteristic pattern of a desired syllable as a characteristic standard pattern, and reading out a waveform in a temporary storage memory corresponding to the syllable instructed by the instructing means. Reproduction output means for outputting a sound, execution instruction means for instructing execution of registration of a feature pattern of a syllable instructed by the instruction means after reproduction by the reproduction output means, and continuous voice based on the instruction of the execution instruction means After the characteristic pattern of the temporary storage memory obtained by the analysis is provided with the second memory registered as the characteristic standard pattern, the desired syllable in the phrase input in the recognition mode is displayed and corrected for confirmation. A Japanese voice input device characterized in that this feature pattern can be registered as a feature standard pattern.