JPH11282490A - Speech recognition device and storage medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To shorten a time required for recognition by reducing a necessary calculation amount before a recognition result is obtained based on a feature parameter obtained by analyzing a speech signal. SOLUTION: In a standard pattern group storage part 20, plural groups which have collected mutually approximates feature parameters from 101 well- known standard patterns, and representative standard patterns using specified standard patterns selected from standard patterns in each group respectively as the standard patterns representing the groups are associated with each other and stored. The representative standard patterns 16a stored in a standard pattern group storage part 20 is stored in a representative standard pattern storage part 16a. And, a comparison recognition part 18 selects the representative standard patterns approximated with the feature parameters extracted by an analysis part 14 from the representative standard pattern storage part 16, and a control unit 22 selects a group corresponding to the selected representative standard pattern from a standard pattern group storage part 20 and outputs it.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition device for recognizing speech and a storage medium storing a computer program for the speech recognition device to perform speech recognition. This can reduce the amount of calculation when performing recognition.

[0002]

2. Description of the Related Art Conventionally, as the above-mentioned speech recognition apparatus, for example, one having a configuration shown in FIG. 5 is known. The input unit 12 provided in the voice recognition device 40 inputs a voice signal to be subjected to voice recognition, and the analysis unit 14 converts the voice signal output from the input unit 12 into a predetermined period (for example, 4 to 20).
msec) to extract feature parameters.
As the characteristic parameter, a parameter extracted from a spectrum included in a voice by a band-pass filter bank is used. Further, the speech recognition device 40 includes a standard pattern memory 42 for storing a standard pattern which is a standard type of feature parameters of 101 single syllables.
In the case of the specific speaker recognition device, the standard pattern is a standard pattern of the characteristic parameters of each single syllable corresponding to the speaker. This is a standard feature parameter of each extracted syllable.

The characteristic parameters analyzed by the analyzing unit 14 are supplied to a comparison recognizing unit 44. The comparison recognizing unit 44 stores the supplied characteristic parameters in a standard pattern stored in a standard pattern memory 42. Is performed for all of the 101 standard parameters stored in the standard pattern memory 42. As a result of the matching calculation processing, the closest approximation is output to the control unit 46 as the first candidate, and the successive approximation is output to the control unit 46 as the next candidate, and then displayed on the display unit 26 such as a CRT or LCD. When a desired recognition result is found from the candidates displayed on the display unit 26, the cursor on the display unit 26 is moved with the keyboard 28 to select the desired recognition result. After the selected recognition result is output to the control unit 46,
Is output to the data processing unit 48.

[0004]

However, the conventional single syllable recognition device 40 performs a matching calculation process for comparing the feature parameters extracted by the analysis unit 14 with the standard patterns stored in the standard pattern memory 42. Since the calculation is performed for all of the 101 standard patterns stored in the standard pattern memory 42, the amount of calculation of the matching calculation processing becomes extremely large, and there is a problem that it takes time.

Accordingly, the present invention provides a speech recognition apparatus and a storage medium which can reduce the amount of calculation required to produce a recognition result based on a characteristic parameter obtained by analyzing a speech signal, thereby shortening the time required for recognition. It is intended to be realized.

[0006]

According to the present invention, in order to achieve the above object, according to the first aspect of the present invention, a feature parameter is extracted by analyzing a monosyllable of a voice, and the extracted feature parameter is extracted. A storage means for storing a group of characteristic parameters by collecting similar parameters in each group, and storing representative parameters generated as parameters representing the group in each group based on characteristic parameters in the group. Voice input means for inputting voice, and analyzing a single syllable of the voice input by the voice input means to extract a characteristic parameter of the single syllable, and representing a representative parameter approximating the extracted characteristic parameter to And output means for selecting and outputting from the storage means.

According to a second aspect of the present invention, in the speech recognition apparatus according to the first aspect, the storage unit stores the representative parameter and the contents of a group represented by the representative parameter in association with each other. The output unit analyzes a single syllable of the voice input by the voice input unit, extracts a characteristic parameter of the single syllable, and associates the extracted characteristic parameter with a representative parameter that approximates the extracted characteristic parameter. The technical means of reading out the contents of the group in question from the storage means and outputting it.

According to a third aspect of the present invention, in the speech recognition apparatus according to the first aspect, the output unit analyzes a single syllable of the voice input by the voice input unit, and analyzes the characteristic parameter of the single syllable. And a technical means of selecting and outputting, from the storage means, a predetermined high-order number having a high degree of approximation among representative parameters approximated to the extracted characteristic parameters.

According to a fourth aspect of the present invention, in the speech recognition apparatus according to the second aspect, the output unit analyzes a single syllable of the voice input by the voice input unit, and analyzes the characteristic parameter of the single syllable. And, from among the representative parameters that are similar to the extracted characteristic parameters, a high-order predetermined number having a high degree of approximation is selected from the storage unit, and the selected number is associated with the selected predetermined number of representative parameters. The technical means of outputting the contents of a certain group is adopted.

According to a fifth aspect of the present invention, in the voice recognition apparatus according to the fourth aspect, the output unit analyzes a single syllable of the voice input by the voice input unit, and analyzes the characteristic parameter of the single syllable. And, from among the representative parameters that are similar to the extracted characteristic parameters, a higher-order predetermined number having a higher degree of approximation is selected from the storage unit, and further, the selected parameters are associated with the selected predetermined number of representative parameters. A technical means is employed in which, from among the feature parameters in the group, a predetermined number of higher-ranked features having a high degree of approximation is selected, and the selected predetermined number of upper-ranked feature parameters are output.

According to a sixth aspect of the present invention, in the speech recognition apparatus according to the second or fourth or fifth aspect, the storage means stores the feature parameters in the group in association with the utterance frequency. The output means employs technical means for outputting the feature parameters in descending order of utterance frequency.

According to a seventh aspect of the present invention, in the speech recognition apparatus according to any one of the first to sixth aspects, the representative parameter is a predetermined characteristic parameter selected from the group. Adopt technical means.

According to the present invention, a feature parameter is extracted by analyzing a single syllable of a voice, and similarities among the extracted feature parameters are collected to obtain a feature parameter. A group is created, and in each group, based on the characteristic parameters in the group, a storage area storing a representative parameter created as a parameter representative of the group and a voice are input, and a monosyllable of the input voice is input. A technical means of a storage medium storing a computer program for analyzing and extracting characteristic parameters of the single syllable and selecting and outputting representative parameters similar to the extracted characteristic parameters from the storage means is employed. .

According to a ninth aspect of the present invention, in the storage medium according to the eighth aspect, the storage area stores the representative parameter and the contents of a group represented by the representative parameter in association with each other. The computer program analyzes a single syllable of the voice input by the voice input unit, extracts a characteristic parameter of the single syllable, and associates the extracted characteristic parameter with a representative parameter that approximates the extracted characteristic parameter. A technical means is employed for reading out and outputting the contents of a certain group from the storage means.

[0015]

According to the first to seventh aspects of the present invention, the storage means analyzes a monosyllable of a voice to extract a characteristic parameter and approximates the extracted characteristic parameter. A group of feature parameters is created by gathering each other, and a representative parameter created as a parameter representative of the group is stored in each group based on the feature parameters in the group. A single syllable of the voice input by the input means is analyzed to extract a characteristic parameter of the single syllable, and a representative parameter similar to the extracted characteristic parameter is selected and output from the storage means. That is, in order to select, from the storage unit, a feature parameter that is similar to the extracted feature parameter by analyzing the audio signal, whether or not all the feature parameters stored in the storage unit are approximated Instead of judging, approximate ones are gathered and grouped, and an approximate one can be selected and output from representative parameters representing the group. Therefore, according to the first to seventh aspects of the present invention, the amount of calculation is smaller than that of the speech recognition device that determines whether or not all the feature parameters stored in the storage unit are approximated. It is possible to realize a speech recognition device capable of reducing the time required for recognition by reducing the number.

In particular, in the second aspect of the present invention, the storage unit stores the representative parameter and the contents of the group represented by the representative parameter in association with each other, and the output unit outputs the voice input. Means for analyzing a single syllable of the input voice and extracting characteristic parameters of the single syllable, and reading out from the storage means the contents of the group associated with the extracted representative parameter and a representative parameter that approximates the characteristic parameter; Output. That is, since the content of the group associated with the representative parameter can be output, the content of the group can be displayed when the output recognition result is displayed.

According to the third aspect of the present invention, the output unit analyzes a single syllable of the voice input by the voice input unit, extracts a characteristic parameter of the single syllable, and extracts the extracted characteristic. Among the representative parameters that are similar to the parameters, a high-order predetermined number having a high degree of approximation is selected from the storage means and output. In other words, not all the representative parameters to be approximated are selected and output, but only the upper predetermined number having a high degree of approximation is selected and output, so that the time for outputting the recognition result can be reduced and the recognition result is displayed. Since the number of displayed cases is reduced, it is easy to find a desired recognition result.

Further, in the invention described in claim 4, the output means analyzes a single syllable of the voice input by the voice input means, extracts a characteristic parameter of the single syllable, and extracts the extracted characteristic. Among the representative parameters that are close to the parameters, a predetermined number with a high degree of approximation is selected from the storage unit, and the contents of the group associated with the selected representative parameter are output. In other words, the contents of the groups associated with the predetermined number of representative parameters having a high degree of approximation can be output. Can be displayed.

In the invention described in claim 5, the output means analyzes a single syllable of the voice input by the voice input means, extracts a characteristic parameter of the single syllable, and extracts the extracted characteristic. Among the representative parameters similar to the parameters, a higher-order predetermined number having a higher degree of approximation is selected from the storage means, and further, among the feature parameters in the group respectively associated with the selected predetermined number of the representative parameters, A predetermined upper-ranked number having a high degree of approximation is selected, and the selected upper-ranked predetermined number of characteristic parameters are output. That is, instead of outputting all of the feature parameters in the upper predetermined number of groups having a higher degree of approximation, selecting and outputting the upper predetermined number of feature parameters having a higher degree of approximation among the characteristic parameters in the group , A feature parameter with a low degree of approximation within a group can be prevented from being output. Therefore, when the recognition result is displayed, feature parameters that are not very similar are not displayed.
It becomes easier to find a desired recognition result.

Further, in the invention according to claim 6, the storage means stores the characteristic parameter in the group in association with the utterance frequency, and the output means outputs the group when outputting the group. Are output in descending order of utterance frequency. In other words, the feature parameters in the group are not output randomly, but are output in descending order of utterance frequency.
Since the probability that the recognition result desired to be selected becomes higher is higher, the desired recognition result can be easily found.

In the invention according to claim 7, technical means is employed in which the representative parameter is a predetermined characteristic parameter selected from within a group. That is,
As a method of creating a representative parameter, based on the feature parameters in the group, a new feature parameter is created, the created feature parameter is used as a representative parameter, and a desired feature parameter in the group is selected. There is a method of using the selected characteristic parameter as a representative parameter. According to the latter method, that is, according to the invention of claim 7, it is not necessary to create a new characteristic parameter unlike the former method. The conventional 101 single syllable patterns described above can be applied as they are.

According to the present invention, a feature parameter is extracted by analyzing a single syllable of a voice, and similar features among the extracted feature parameters are collected to obtain a feature parameter. Create a group,
In each group, based on the characteristic parameters in the group, a storage area storing a representative parameter created as a parameter representative of the group, a voice is input, a single syllable of the input voice is analyzed, and the unit is analyzed. Since the computer program is configured to extract a syllable feature parameter and select and output a representative parameter similar to the extracted feature parameter from the storage area, the computer program is stored. Thus, the speech recognition device according to claim 1 can be realized. In other words, the speech recognition device is controlled by a CPU incorporated in the speech recognition device or a computer connected to the speech recognition device, as described in an embodiment of the invention described later. By providing a storage unit as a storage medium in a speech recognition device, or by installing a computer program stored in the storage medium into a computer, a speech recognition device capable of reducing the amount of calculation and shortening the time required for recognition. Because it can be realized.

[0023] In particular, in the ninth aspect of the present invention, the storage area stores a representative parameter and a content of a group represented by the representative parameter in association with each other. Means for analyzing a single syllable of the input speech and extracting characteristic parameters of the single syllable, and reading out the contents of a group associated with a representative parameter approximating the extracted characteristic parameter from a storage area; Since it is for outputting, it is possible to output the contents of the group associated with the representative parameter, and a speech recognition device that can display the contents of the group when displaying the output recognition result. realizable.

[0024]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the speech recognition apparatus of the present invention will be described below with reference to the drawings. FIG. 1 is an explanatory diagram showing a schematic configuration of the speech recognition device of the present embodiment by blocks. The same components as those in the related art are denoted by the same reference numerals, and description thereof is omitted. As shown in FIG. 1, the speech recognition apparatus 10 of the present embodiment includes an input unit 12, a representative standard pattern storage unit 16, an analysis unit 14, a comparison recognition unit 18, Standard pattern group storage unit 20
, Control unit 22, data processing unit 24, display unit 26
And a keyboard 28. Note that the analysis unit 14, the comparison recognition unit 18, and the control unit 22 function as output means of the present invention.

The input unit 12 includes a microphone, an amplifier,
It is composed of an A / D converter and a filter, etc.
The analysis unit 14 includes a band-pass filter bank and the like. A CPU is used as the comparison recognition unit 18, and a storage medium such as a ROM, an EEPROM, and a RAM is used as the representative standard pattern storage unit 16 and the standard pattern group storage unit 20. The display unit 26 includes, for example, a monitor television, an LCD (liquid crystal display device), and the like. Here, the storage contents of the representative standard pattern storage unit 16 and the standard pattern group storage unit 20 will be described with reference to FIG. FIG. 2A is an explanatory diagram showing the storage contents of the standard pattern group storage unit 20, and FIG. 2B is an explanatory diagram showing the storage contents of the representative standard pattern storage unit 16. Note that the storage contents of the standard pattern group storage unit 20 and the representative standard pattern storage unit 16 are actually data of a standard pattern which is a standard type of a characteristic parameter of a single syllable, but the description is easy to understand. For this reason, in FIG. 2, the storage contents are indicated by single syllables corresponding to the standard pattern. As shown in FIG. 2A, the standard pattern group storage unit 20 stores a group 20a composed of four types of standard patterns of 50 tones, a murmur, a voiced sound, and a semi-voiced sound, and a group 20a.
And n standard patterns indicating 50 sounds from the standard pattern group in the table are stored in association with a standard pattern (hereinafter, referred to as a representative standard pattern) 16a selected as a standard pattern representing the group. For example, the first group contains four characters, “ka”, “ki”, “ga”, and “gyu”.
A group 20a composed of a plurality of standard patterns and “ka”, which is the representative standard pattern 16a of the group 20a, are stored in association with each other. As shown in FIG. 2B, the representative standard pattern storage unit 16 stores the n representative standard patterns 16 stored in the standard pattern group storage unit 20.
a is stored. Note that the representative standard pattern corresponds to the representative parameter of the present invention.

Next, the flow of processing when the speech recognition apparatus 10 of this embodiment performs speech recognition will be described with reference to the flowchart of FIG. The analysis unit 14, comparison / recognition unit 18, and control unit 22 shown below function mainly when a CPU (not shown) executes a computer program. First, when the user of the voice recognition device 10 generates a single syllable toward a microphone or the like provided in the input unit 12 (step 2), the generated single syllable is input to the A / It is converted into a digital audio signal by a D converter and a filter. Subsequently, the analysis unit 14 sets the input unit 1
The audio signal output from the second is analyzed at every predetermined cycle into a time change of a frequency spectrum to extract a characteristic parameter (step 4). Subsequently, the comparison recognition unit 18 compares the characteristic parameters extracted by the analysis unit 14 with the representative standard patterns 16 stored in the representative standard pattern storage unit 16.
a are sequentially compared (step 6).

The comparison / recognition unit 18 selects the representative standard pattern closest to the representative standard pattern 16a obtained by comparison as the first candidate, and selects the representative standard pattern sequentially approximated as the next candidate. Then, the selected representative standard pattern is output to the control unit 22 (step 10). Subsequently, the control unit 22 selects the group 20a associated with the input representative standard pattern, outputs data indicating the standard pattern that is the contents of the selected group to the display unit 26, and outputs the data to the display unit 26. Displays a single syllable indicated by the input data (step 12). For example, when the user utters “Kyu”, the comparison / recognition unit 18 sets “ka” as the first candidate in the representative standard pattern storage unit 16 shown in FIG.
The first to fourth candidates are selected, such as "ta" as a candidate (step 8), and the control unit 22 selects the first to fourth candidates.
As a group associated with the representative standard pattern up to the fourth candidate, the standard pattern group storage unit 2 shown in FIG.
Four groups are selected from 0, and data indicating a standard pattern, which is the contents of each of the four groups, is output to the display unit 26. Then, the four selected groups are displayed on the display unit 26 as shown in FIG. 4 for explaining the recognition result (step 12).

After the user finds a desired recognition result from the recognition results displayed on the display unit 26, the user operates the keyboard 28 to move the cursor on the screen to the desired recognition result. (Step 14). Subsequently, the selected recognition result is output to the data processing unit 24 after being output to the control unit 22 (step 1).
6). In addition, as shown in FIG. 2A, a number may be assigned to the standard pattern and the representative standard pattern, and the recognition result may be displayed on the display unit 26 together with the number as shown in FIG. it can. According to this configuration, the user:
By directly entering the number using the numeric keypad, etc.
A desired recognition result can be directly selected. Further, the data output from the control unit 22 may be output to a storage medium (not shown) and stored in the storage medium.

As described above, by using the speech recognition apparatus 10 of the present embodiment, standard patterns having similar characteristic parameters are grouped, and a representative standard pattern representing the group is defined for each group. By setting, the number of comparison targets is reduced, and the determination as to whether or not it is close to the characteristic parameter extracted by the analysis of the audio signal is performed only for the representative standard pattern 16a with a small number. As described above, the amount of calculation can be considerably reduced and the time required for recognition can be reduced as compared with the case where the above determination is made for all 101 standard parameters. For example, when the number of representative standard patterns stored in the representative standard pattern storage unit 16 is 20, the number of matching processes of the comparison recognition unit 18 is 20, which can be reduced to about 1/5 of the conventional calculation amount.

In the above embodiment, the comparison recognition unit 18
Is used to select up to four higher-order representative standard patterns that are determined to be approximated from the representative standard pattern storage unit 16, but the number is not limited to four, and for example, two or three are selected. Such a configuration may be adopted, and further, a selection may be made without any particular limitation. With this configuration, the number of selection of the representative standard pattern can be reduced, so that the processing time of the comparison recognition unit 18 and the control unit 22 can be reduced, and the number of recognition results displayed on the display unit 26 can be reduced. This has the effect of making it easier to find the recognition result. This configuration corresponds to the third or fourth aspect of the present invention.

Further, a configuration is provided for selecting a standard pattern that is similar to the characteristic parameter extracted by analyzing the audio signal from the group selected by the control unit 22 from the standard pattern group storage unit 20. A configuration in which the standard pattern and the representative standard pattern are displayed on the display unit 26 may be used. That is, instead of displaying the predetermined number of representative standard patterns and all of the standard patterns in the group, a standard pattern that is close to the standard patterns in the group is selected and displayed together with the representative standard pattern. Therefore, it is possible not to output a non-approximate standard pattern in the group. Therefore, when the recognition result is displayed, a feature parameter that is not so approximated is not displayed, so that the recognition result can be easily found. This configuration corresponds to the fifth aspect of the present invention.

Further, the control unit 22 may be configured to output recognition results in descending order of utterance frequency. For example, calculate the utterance frequency of a single syllable in advance,
The calculation result is stored in the standard pattern group storage unit 20 as data indicating the utterance frequency in association with the standard pattern.
Then, when the control unit 22 outputs the group selected from the standard pattern group storage unit 20 to the display unit 26, based on the data indicating the utterance frequency associated with the standard pattern in the group to be output, The standard patterns are output to the display unit 26 in descending order of the utterance frequency. With this configuration, recognition results are output in descending order of utterance frequency.
When the recognition result is displayed, the probability that the recognition result desired to be selected becomes higher is increased, and there is an effect that the desired recognition result is easily found. In addition, this configuration corresponds to claim 6
Corresponds to the invention described in (1).

Further, in the above embodiment, matching with a standard pattern is used as a method for recognizing a single syllable. However, a recognition method using a hidden Markov model, a neural network, or the like can be used. Further, in the above-described embodiment, a predetermined standard pattern is selected from among standard patterns in a group in which characteristic parameter approximations are collected from among 101 known standard patterns,
The selected standard pattern is set as the representative standard pattern, but it is divided into single syllable groups whose characteristic parameters are similar, and learning is performed from the single syllable groups in the group to generate representative characteristic parameters. , Can be used as a representative standard pattern. When using DP matching as a method for recognizing a single syllable, the learning is performed by using an average value of feature parameters in a group or extracting a common part of feature parameters. When a hidden Markov model is used, Baum-Welch (expec
tation-maximization-algor
It learns using (i.

The computer program for the voice recognition device 10 of the above embodiment to perform voice recognition is provided at a predetermined location in the voice recognition device 10 as a storage medium such as a ROM, or the computer program is stored on a CD-ROM. It is provided by storing it in a ROM, a floppy disk, or the like, and installing them using a reading device provided in the device. In this case, the ROM, CD-ROM, floppy disk, or the like functions as the storage medium according to the third or ninth aspect. Furthermore, a computer program can be read from an external information processing device via a wired or wireless communication unit and operated.

Incidentally, the representative standard pattern storage section 16 and the standard pattern group storage section 20 in the above embodiment correspond to the storage means of the present invention, and the storage contents of the representative standard pattern storage section 16 and the standard pattern group storage section 20 are as follows. If they are stored as a part of a computer program, an area for storing them corresponds to the storage area according to claim 8 or 9.

[0036]

As described above, claims 1 to 7 are as described above.
According to the invention described in the above, while analyzing a single syllable of the voice to extract the feature parameters, among the extracted feature parameters that are similar to each other are collected to create a group of feature parameters, In the group, based on the characteristic parameters in the group, storage means for storing a representative parameter created as a parameter representative of the group, and a single syllable of the voice input by the voice input means for inputting the voice are analyzed. Output means for extracting a characteristic parameter of a single syllable and selecting and outputting a representative parameter approximating the extracted characteristic parameter from the storage means, so that all characteristic parameters stored in the storage means are provided. When the amount of calculation is smaller than that of a speech recognition device that determines whether or not Can be realized speech recognition apparatus can be shortened.

In particular, according to the second aspect of the present invention, the storage means stores the representative parameter and the contents of the group represented by the representative parameter in association with each other. A single syllable of the voice input by the voice input unit is analyzed to extract the characteristic parameter of the single syllable, and the contents of the group associated with the extracted representative parameter and the representative parameter that is approximated are stored from the storage unit. Since the data is read and output, the content of the group can be displayed when the output recognition result is displayed.

According to the third aspect of the present invention, the output means analyzes a single syllable of the voice input by the voice input means, extracts a characteristic parameter of the single syllable, and extracts the characteristic parameter of the single syllable. Among the representative parameters that are similar to the characteristic parameters, a predetermined number with a high degree of approximation is selected and output from the storage unit, so that the time required to output the recognition result can be reduced, and the display when displaying the recognition result can be shortened. Since the number is reduced, a desired recognition result can be easily found.

Further, according to the invention described in claim 4,
The output means analyzes a single syllable of the voice input by the voice input means, extracts a characteristic parameter of the single syllable, and, among representative parameters approximated to the extracted characteristic parameter, Is selected from the storage means, and the contents of the groups respectively associated with the selected representative parameters are output. Therefore, when displaying the output recognition results, the upper predetermined number of groups having a high degree of similarity are displayed. Can be displayed.

According to the fifth aspect of the present invention, the output means analyzes a single syllable of the voice input by the voice input means, extracts the characteristic parameter of the single syllable, and extracts the characteristic parameter of the single syllable. Of the representative parameters approximated by the selected characteristic parameter, a high-order predetermined number having a high degree of approximation is selected from the storage means, and further, the characteristic parameters in the group respectively associated with the selected predetermined number of the representative parameters are selected. Of these, a higher-order predetermined number having a high degree of approximation is selected, and the selected higher-order predetermined number of feature parameters are output. When the recognition result is displayed, feature parameters that are not very similar are not displayed. Makes it easier to find the recognition result.

Further, according to the invention described in claim 6,
The storage unit stores the feature parameters in the group in association with the utterance frequency, and the output unit outputs the feature parameters in descending order of the utterance frequency. The higher the probability that the recognition result to be performed is higher, the easier it is to select the recognition result. With this configuration, since the recognition results are output in descending order of utterance frequency, when the recognition results are displayed, the probability that the recognition result desired to be selected becomes higher is higher, and the desired recognition result is more easily found. It has the effect of becoming.

According to the seventh aspect of the present invention, a technical means is adopted in which the representative parameter is a predetermined characteristic parameter selected from within a group. In other words, as a method of creating a representative parameter, a new feature parameter is created based on the feature parameters in the group, and the created feature parameter is used as a representative parameter, and a desired feature parameter in the group is selected. Then, there is a method in which the selected characteristic parameter is used as a representative parameter.
According to the invention described in (1), unlike the former method, it is not necessary to create a new feature parameter. For example, the above-mentioned conventional 101 single syllable patterns can be applied as they are.

According to the eighth aspect of the present invention,
A feature parameter is extracted by analyzing a monosyllable of a voice, and a group of feature parameters is created by collecting approximations among the extracted feature parameters, and a group of feature parameters is created for each group. Based on the storage area storing a representative parameter created as a parameter representing the group, and input a voice, and analyze a single syllable of the input voice to extract the characteristic parameter of the single syllable, Since the configuration is a storage medium in which a computer program for selecting and outputting a representative parameter similar to the extracted characteristic parameter from the storage area is stored, the storage medium is provided as a storage unit in the speech recognition device, or The computer program stored in the storage medium, It can be realized speech recognition apparatus which can shorten the time required for the recognition by reducing the amount of calculation by installing a computer connected to a voice recognition device.

In particular, according to the ninth aspect of the present invention, the storage area stores the representative parameter and the contents of the group represented by the representative parameter in association with each other. A single syllable of the voice input by the voice input means is analyzed to extract the characteristic parameter of the single syllable, and the contents of the group corresponding to the extracted representative parameter and the representative parameter approximated from the storage area. Since it is for reading and outputting, it is possible to output the contents of the group associated with the representative parameter, and to display the contents of the group when displaying the output recognition result. The device can be realized.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram showing a schematic configuration of a speech recognition device according to an embodiment of the present invention by blocks.

FIG. 2A is an explanatory diagram showing storage contents of a standard pattern group storage unit 20; FIG. 2B is an explanatory diagram showing storage contents of a representative standard pattern storage unit 16;

FIG. 3 is a flowchart showing a processing flow of the voice recognition device according to the embodiment of the present invention.

FIG. 4 is an explanatory diagram showing a recognition result displayed on a display unit 26;

FIG. 5 is an explanatory diagram showing a schematic configuration of a conventional voice recognition device by blocks.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 Speech recognition apparatus 12 Input part 14 Analysis part 16 Representative standard pattern storage part 16a Representative standard pattern 18 Comparative recognition part 20 Standard pattern group storage part 20a Group 22 Control part 24 Data processing part 26 Display part 28 Keyboard

Claims (9)

[Claims] 1. A feature parameter is extracted by analyzing a monosyllable of a speech, and a group of feature parameters is created by collecting approximations among the extracted feature parameters. Based on the characteristic parameters in the storage unit, storage means for storing representative parameters created as parameters representing the group, voice input means for inputting voice, and monosyllables of voice input by the voice input means are analyzed. Output means for extracting a characteristic parameter of the single syllable and selecting and outputting a representative parameter similar to the extracted characteristic parameter from the storage means. . 2. The storage unit stores the representative parameter and the contents of a group represented by the representative parameter in association with each other, and the output unit outputs a voice input by the voice input unit. Analyzing a single syllable to extract the characteristic parameter of the single syllable, and reading and outputting from the storage means the contents of a group associated with a representative parameter that is similar to the extracted characteristic parameter. The voice recognition device according to claim 1. 3. The output means analyzes a single syllable of the voice input by the voice input means to extract a characteristic parameter of the single syllable, and includes a representative parameter which is similar to the extracted characteristic parameter. 2. The speech recognition apparatus according to claim 1, wherein a predetermined number having a higher degree of approximation is selected and output from the storage unit. 4. The output means analyzes a single syllable of the voice input by the voice input means to extract a characteristic parameter of the single syllable, and includes a representative parameter which is similar to the extracted characteristic parameter. 3. The method according to claim 2, wherein a predetermined number having a high degree of approximation is selected from the storage unit, and the contents of the group respectively associated with the selected predetermined number of representative parameters are output. Voice recognition device. 5. The output means analyzes a single syllable of the voice input by the voice input means to extract a characteristic parameter of the single syllable, and includes a representative parameter which is similar to the extracted characteristic parameter. , Selecting a predetermined number with a high degree of similarity from the storage means, and further selecting a predetermined number with a high degree of similarity among the feature parameters in the group respectively associated with the selected predetermined number of representative parameters. 5. The speech recognition apparatus according to claim 4, wherein the number is selected, and a selected predetermined number of characteristic parameters of the selected higher order are output. 6. The storage unit stores feature parameters in the group in association with utterance frequencies, and the output unit outputs the feature parameters in descending order of utterance frequency. The speech recognition device according to claim 2 or claim 4 or claim 5. 7. The speech recognition apparatus according to claim 1, wherein the representative parameter is a predetermined characteristic parameter selected from the group. 8. A monosyllable of a voice is analyzed to extract feature parameters, and a group of feature parameters is created by collecting approximations among the extracted feature parameters. Based on the characteristic parameters in the group, a storage area storing representative parameters created as parameters representative of the group, a voice is input, a single syllable of the input voice is analyzed, and the A storage medium storing a computer program for extracting characteristic parameters and selecting and outputting a representative parameter similar to the extracted characteristic parameters from the storage means. 9. The storage area stores the representative parameter and the contents of a group represented by the representative parameter in association with each other, and the computer program stores a voice of the voice input by the voice input unit. This is for analyzing a single syllable and extracting a characteristic parameter of the single syllable, and reading and outputting the contents of a group associated with a representative parameter approximated to the extracted characteristic parameter from the storage means. 9. The storage medium according to claim 8, wherein: