JPH07303053A - Area discriminator and speech recognizing device - Google Patents

Abstract

PURPOSE:To provide a high speech recognition ratio in spite of a working area without applying operating burden on a user. CONSTITUTION:This discriminator is provided with a carrier frequency storage means 110 which stores the carrier frequency of prescribed radio or television broadcasting at every area, and a broadcasting wave decision means 111 which judges the existence of a broadcasting wave of stored carrier frequency by the presence/absence of a power. Moreover, the discriminator is provided with an area estimating means 109 which estimates an area to be used corresponding to the broadcasting wave of carrier frequency whose existence is judged, and every area reference pattern storage means 107, 108 which store a reference pattern at every area and output the reference pattern at every area estimated by the area estimating means 109. Speech recognition using the reference pattern in accordance with the working area can be conducted.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an area determination device and a voice recognition device.

[0002]

2. Description of the Related Art Among various devices, their functions and processing accuracy are affected by regional characteristics. For example, the voice recognition device is intended to process human voiced speech,
Since the voiced speech is affected by the dialect and accent of each region, the recognition accuracy of the speech recognition device is ultimately affected by the regional characteristics.

Therefore, as a voice recognition device, a device considering the influence of regional characteristics has already been proposed (for example, Japanese Patent Laid-Open No. 60-73595 and Japanese Patent Laid-Open No. 61-70594).
Issue).

An example of a conventional speech recognition apparatus considering the influence of regional characteristics will be described with reference to FIG. In FIG. 2, the voice recognition device includes a control unit 201, a microphone 202 for voice input, a pre-processing unit 203 including a filter that passes only the frequency band of voice, and an analog / digital conversion unit that converts voice into a digital value. (A / D converter) 20
4. An analysis unit 205 for extracting a voice feature from the converted data, a standard pattern 207 created in advance, and a matching unit 206 for performing pattern matching with the feature parameter of the input voice and the standard pattern.

With such a configuration, first, the microphone 2
The voice input from 02 is limited to the signal of only the voice frequency band in the pre-processing 203. The signal is sent to the control unit 2
The analog / digital conversion start signal from 01 causes the analog / digital conversion unit 204 to perform analog / digital conversion. The analysis unit 205 calculates characteristic amount parameters such as a power spectrum and LPC cepstral based on the digital data. Then, the matching unit 206 recognizes the input voice by comparing the standard pattern 207 created based on all domestic dialects with the feature amount parameter in advance by the same method and obtaining the pattern similarity.

[0006]

By the way, in the case of this speech recognition apparatus, the created standard pattern 207 is selected as the pattern having the maximum similarity with the feature amount parameter of any dialect. However, there is a large difference between dialects, such as vowels being erased in the Kansai region while vowels are erased in the Kanto region, and it cannot be said that some dialects have an appropriate pattern. . In other words, when using a standard pattern that reflects all the differences in accents and intonations of dialects in this way, if you limit it to only one dialect, it will be a pattern that blurs its characteristics. For strong words,
There was a problem that there was a high probability that it would be misrecognized.

Therefore, a voice recognition device has already been proposed in which a standard pattern is created and stored for each region (for each dialect) and the standard pattern of the region designated by key input is used for recognition. However, since the user has to specify the area, there is room for improvement in operability, and a device for automatically determining the area is required.

As for the voice recognition device, there is one that automatically identifies the region by detecting the region attribute of all or a part of the recognized words. Because of this, the accuracy of identifying the area is low.

[0009]

In order to solve such a problem, the first present invention provides (1) carrier frequency storage means for storing a carrier frequency of a predetermined radio broadcast or television broadcast for each region, (2) The existence of the broadcast wave of the carrier frequency stored in this carrier frequency storage means
An area determination device comprising a broadcast wave determination means for determining whether or not there is power, and (3) an area estimation means for estimating an area according to a broadcast wave of a carrier frequency whose presence is determined by this broadcast wave determination means. provide.

In the second aspect of the present invention, standard patterns relating to linguistic features included in voice are stored in advance,
The following means are provided in a voice recognition device that recognizes a voice based on the similarity between the input voice and the standard pattern.

That is, (1) carrier frequency storage means for storing a carrier frequency of a predetermined radio broadcast or television broadcast for each region, and (2) a broadcast wave of a carrier frequency stored in the carrier frequency storage means. The presence of the broadcast wave determining means to determine the presence or absence of power, (3) in accordance with the broadcast wave of the carrier frequency of the presence is determined by the broadcast wave determining means, the area estimation means to estimate the area used (4) A standard pattern for each area is stored, and the area standard pattern storage means outputs the standard pattern of the area estimated by the area estimation means.

[0012]

Both the first and second aspects of the present invention are based on the fact that the carrier frequencies (channel frequencies for television) of radio broadcasts and television broadcasts that are broadcast in each region are different. That is, the carrier frequency being broadcast is detected, and the region is estimated based on the detected carrier frequency (or a combination thereof).

In the second aspect of the present invention, the standard pattern used for matching is automatically changed in accordance with the region thus estimated to improve the voice recognition accuracy.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A first embodiment of the present invention will be described in detail below with reference to the drawings. FIG. 1 shows a voice recognition device according to the first embodiment.

In FIG. 1, the voice recognition apparatus of this embodiment has a pre-processing unit including a control unit 101 similar to a conventional voice recognition apparatus, a microphone 102 for voice input, a filter for passing only a voice frequency band, and the like. Unit 103, analog / digital conversion unit 104 for converting a voice signal into a digital value, analysis unit 105 for extracting a characteristic parameter of voice from the converted data, rewritable eg EEP
A standard pattern 107 stored in a ROM or the like is provided.

Further, in the voice recognition apparatus of this embodiment, a matching unit 106 that performs pattern matching between the characteristic parameter of the input voice and the standard pattern, a pattern different from the standard pattern in each region (called a change pattern).
Change pattern storage unit 10 in which information is compressed and stored
8. Carrier frequency storage unit 110 for storing carrier frequencies of FM broadcasting stations nationwide, carrier signal presence / absence determination unit 11 for determining presence / absence of power at carrier frequencies
1, an antenna 112 for receiving broadcast waves, and a use area estimation unit 109 for estimating an area where the voice recognition device is used based on the results of the carrier signal presence / absence determination unit 111.

In the above configuration, when the voice recognition device is activated, the control unit 101 first transmits a determination start signal to the carrier signal presence / absence determination unit 111.

As a result, the carrier signal presence / absence determining unit 11
1 indicates that the carrier frequency of the FM broadcasting station is set as the resonance frequency from the carrier frequency storage unit 110, as shown in FIG.
The presence or absence of power is determined for each station. The determination result of the presence or absence of the power is sent to the usage area estimation unit 109 and the usage area is estimated.

In FIG. 3, for example, "FM
Since the power of “Gunma” has five levels and the power of “Natsukhuive” has two levels, the use area estimation unit 109 estimates that the area is used in Gunma prefecture. The use area estimation unit 109 associates, for example, an area in which a change pattern for the standard pattern as shown in FIG. 4 is prepared with a combination of standard carrier frequencies and power stages for determining the area. Data is built in, and the region is estimated by matching with the determination result of the power presence / absence from the carrier signal presence / absence determining unit 111.

The use area estimation unit 109 sends a data compression / decompression signal of the Gunma prefecture change pattern to the change pattern storage unit 108. As a result, the change pattern storage unit 108
Decompresses the change pattern for Gunma prefecture and transfers the change pattern of the change pattern in the standard pattern 107.

Next, the use area estimation unit 109 transmits a standard pattern change end signal to the control unit 101, which
Voice input is enabled.

After the voice is input, the voice input from the microphone 102 is limited to the signal of only the frequency band of the voice by the preprocessing unit 103, as in the conventional case. The signal is analog / digital converted by the analog / digital converter 104 in response to an analog / digital conversion start signal from the controller 101. The analysis unit 105 calculates a characteristic amount, for example, a characteristic amount such as a power spectrum or an LPC cepstrum, based on the digital data. Matching unit 106
Then, the input voice is recognized by comparing the above-mentioned Gunma prefecture standard pattern with the pattern similarity by, for example, DP matching.

At present, there are about one FM broadcasting station in each prefecture, and by storing the carrier frequencies of FM broadcasting stations nationwide in the carrier frequency storage section 110 as described above, The area where the device is used can be estimated.

According to the first embodiment, the carrier frequencies of FM broadcasting stations nationwide are stored, the area used is estimated based on the presence or absence of the carrier frequency of the FM broadcasting stations, and the standard pattern is changed for each area. With this optimization, even a speaker in an area with a strong dialect can recognize speech with high accuracy.

Further, according to the first embodiment, since the area is automatically estimated, the key input for designating the area is not necessary and the usability is improved. For example, even if a device provided with the voice recognition device as a partial configuration is moved to a different area, it is not necessary to perform the key input for specifying the area again.

Further, according to the first embodiment, instead of storing the standard pattern for each area, only a portion different from the standard pattern common to each area is prepared as a change pattern and each area is prepared. Since the standard pattern is realized and the changed pattern is compressed and stored, the standard pattern can be switched for each region, but the increase in storage capacity can be suppressed to a slight one. it can.

Next, a second embodiment of the present invention will be described with reference to FIG. FIG. 5 shows the voice recognition apparatus of the second embodiment, and the same or corresponding parts as those of the first embodiment described above are designated by the same reference numerals.

In the voice recognition device of the second embodiment, a configuration obtained by analyzing the meaning of the voice recognition result is added to the configuration of the first embodiment. That is, the semantic analysis unit 120,
Standard data 121 for semantic analysis and change data storage unit 12
2 has been added. Here, the semantic analysis unit 120 obtains the meaning by performing, for example, morphological analysis or syntactic analysis on the speech recognition result using the standard data 121 for semantic analysis. The standard data 121 for semantic analysis stores standard data used for semantic analysis such as word information and its part-of-speech information. Standard data used for semantic analysis such as part-of-speech information is stored in a compressed form, and the estimated change data of the area is used in place of the standard data.

The speech recognition apparatus of the second embodiment, to which the above-mentioned new components are added, will not be described in detail when it is activated, but like the first embodiment, first the area estimation operation is executed. To do.

When the use area estimation unit 109 estimates the use area (for example, Gunma prefecture) of the voice recognition device, the change pattern storage unit 108 and the change data storage unit 122 respectively set the change pattern for Gunma prefecture and Gunma prefecture. The data compression / decompression signal of the modification data is transmitted. As a result, the change pattern storage unit 108 decompresses the data about the Gunma prefecture change pattern, saves the transfer destination data of the change pattern in the standard pattern 107 to the unused area, and transfers the change pattern. The storage unit 122 decompresses the data compression for the Gunma prefecture modified data and decompresses the standard data 121.
The transfer destination data of the change data therein is saved in an unused area and the change data is transferred.

Next, the use area estimation unit 109 transmits a change end signal to the control unit 101, whereby the voice input is possible.

Since the voice recognition operation itself is the same as that of the first embodiment, its explanation is omitted. When the speech recognition result is obtained, the semantic analysis unit 120 performs, for example, morphological analysis or syntactic analysis on the speech recognition result based on the semantic analysis standard data 121, a part of which is replaced with the Gunma prefecture change data. And get the meaning.

Therefore, according to the second embodiment, the same effect as that of the first embodiment can be obtained, and further, the meaning of the input voice can be understood with high accuracy.

For example, in the case of the pronunciation sound "voice" (voice recognition result), it means "I" in the Osaka dialect and "you" in the Kagoshima dialect. Further, in the case of the pronunciation sound "voice" (speech recognition result), it means "you" in the Osaka dialect and "I" in the Kagoshima dialect. The device of this embodiment can estimate the area,
Since the standard data for semantic analysis (database) can be changed and optimized according to the estimated area, the accuracy rate of semantic understanding can be improved.

In the above embodiment, the area is estimated based on the carrier frequency of FM broadcasting. However, the area may be estimated based on the carrier frequency of other radio broadcasting. Also, the region may be estimated based on the carrier frequency of the television broadcast, or the region may be estimated by comprehensively judging both carrier frequencies.

Further, the area estimation configuration in the above embodiment can be applied not only to the voice recognition apparatus but also to other apparatuses that require area discrimination. For example, it can be applied to a car navigation system, and can be used in such a form that a map of an estimated area is displayed first when the system is started.

[0037]

As described above, according to the area determining apparatus of the first aspect of the present invention, the area is based on the fact that the carrier frequencies of the radio broadcasts and the television broadcasts broadcast in each area are different. Since the area is estimated, the area can be estimated with high accuracy, and when incorporated into an apparatus or system that requires area information, the processing accuracy of the apparatus or system can be improved.

According to the voice recognition device of the second aspect of the present invention, the region is estimated and matching is performed based on the fact that the carrier frequency of the radio broadcast or the television broadcast which is broadcast in each region is different. Since the standard data is automatically optimized according to the estimated area, it is possible to improve the recognition rate as compared with the conventional method without imposing an operation burden on the user.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a voice recognition device according to a first embodiment.

FIG. 2 is a block diagram showing a conventional voice recognition device.

FIG. 3 is a characteristic curve diagram for explaining an area estimation operation according to the first embodiment.

FIG. 4 is an explanatory diagram of stored data necessary for an area estimation operation according to the first embodiment.

FIG. 5 is a block diagram showing a configuration of a voice recognition device in a second embodiment.

[Explanation of symbols]

101 ... Control unit, 105 ... Analysis unit, 106 ... Matching unit, 107 ... Standard pattern, 108 ... Conversion pattern storage unit, 109 ... Use area estimation unit, 110 ... Carrier frequency storage unit, 111 ... Carrier signal presence / absence determination unit, 120 ... Semantic analysis unit 121 ... Standard data for semantic analysis 122 ... Change data storage unit.

Claims (4)

[Claims] 1. A carrier frequency storage means for storing a carrier frequency of a predetermined radio broadcast or television broadcast for each area, and the existence of a broadcast wave having a carrier frequency stored in the carrier frequency storage means, An area determination device comprising: a broadcast wave determination means for determining whether the area is present or not; and an area estimation means for estimating an area according to a broadcast wave having a carrier frequency of which presence is determined by the broadcast wave determination means. 2. A voice recognition device for pre-storing a standard pattern relating to linguistic features contained in a voice and recognizing a voice based on the similarity between the input voice and the standard pattern, in a predetermined radio broadcast for each region or Carrier frequency storage means for storing the carrier frequency of television broadcasting, broadcast wave determination means for determining the presence of a broadcast wave of the carrier frequency stored in the carrier frequency storage means by the presence or absence of power, and this broadcast wave determination Area estimation means that estimates the area to be used according to the broadcast wave of the carrier frequency whose presence is determined by the means, and the standard pattern for each area are stored, and the area standard estimated by the area estimation means is stored. A voice recognition device comprising: a regional standard pattern storage unit that outputs a pattern. 3. A change in which the area standard pattern storage means stores a common standard pattern storage section that stores a standard pattern that is common to each area, and a change pattern that stores a change pattern for a standard pattern that is unique to each area. A common standard pattern that reads the change pattern stored in the change pattern storage unit according to the pattern storage unit and the area estimated by the area estimation unit, and changes the common standard pattern according to the change pattern. The voice recognition device according to claim 2, wherein the voice recognition device comprises a changing unit. 4. The speech recognition apparatus according to claim 2, wherein the semantic analysis means performs a semantic analysis on the speech recognition result based on the semantic analysis data, and the semantic analysis data for each region. A voice recognition device, which stores the meaning analysis data for each area that stores the meaning analysis data of the area estimated by the area estimation means to the meaning analysis means.