JP6723907B2 - Language recognition system, language recognition method, and language recognition program - Google Patents

Description

The present invention relates to a language recognition system, a language recognition method, and a language recognition program.

Various developments are currently underway regarding a technique for identifying the language of words spoken by a person (eg, Japanese, English, Chinese, Korean, etc.). For example, in Patent Document 1, a speech recognition for language identification is performed in which a predetermined calling operation is performed for a user and a limited pattern of words that may be returned in response to the calling operation is a recognition candidate in the language identification. The engine is listed.

JP, 2004-53825, A

However, in Patent Document 1, although effective in language identification in a specific business such as ATM of a bank, utterances made for recognizing the language are limited, and the application range is narrow.

Therefore, it may be possible to calculate the reliability of each language based on the recognition result output from the utterance input to the speech recognition device of each language and compare the calculated reliability to identify the language. To be However, even with such a method, it is difficult to accurately identify when the utterance is short.

In particular, when a special word is uttered in a specific task, for example, when the speech of a speaker includes a word that is difficult to recognize by a speech recognition device in any language, the calculated language The degree of reliability is low, and as a result, correct language recognition may not be possible. For example, in tourism business, if the language spoken by a foreign tourist includes a place name of a country other than the tourist's home country, it is very difficult to identify the language spoken by the tourist. .. As described above, there is a problem that the language identification becomes difficult depending on the type of the spoken words.

The present invention has been made based on such a background, and an object thereof is to provide a language recognition system, a language recognition method, and a language recognition program for accurately performing language recognition.

One of the present inventions for solving the above-mentioned problem is a language recognition system that includes a processor and a memory and recognizes a language of a voice that is emitted, and includes a voice acquisition unit that acquires a voice and a first language. A first language general translation unit that generates a first language general translation phrase that is a word in the first language corresponding to the acquired voice, based on first language general vocabulary information that is information about the first vocabulary; A first language specialized translation phrase, which is a word in the first language corresponding to the acquired voice, is generated based on first language specialized vocabulary information that is information about a second vocabulary in the first language. A first specialized section calculating unit that extracts a first specialized language translation unit, the generated first language general translation phrase, and a first language corresponding portion that is a mutually corresponding portion between the generated first language specialized translation phrase. And a second language general translation phrase that is a word in the second language corresponding to the acquired voice, based on second language general vocabulary information that is information about the third vocabulary in the second language. A second language specialized language, which is a word in the second language corresponding to the acquired voice, based on a bilingual general translation unit and second language specialized vocabulary information that is information about the fourth vocabulary in the second language. A second-language specialized translation unit that generates a translated phrase, the generated second-language general translated phrase, and a second-language corresponding portion that is a mutually corresponding portion of the generated second-language specialized translated phrase are extracted. And a language identifying unit that identifies the language of the acquired voice based on the extracted first language corresponding part and the extracted second language corresponding part.

According to the present invention, language recognition can be accurately performed.

Problems, configurations, and effects other than those described above will be clarified by the following description of the embodiments.

FIG. 1 is a diagram showing an example of the configuration of a language recognition system 1 according to this embodiment. FIG. 2 is a diagram illustrating an example of the hardware configuration of the service robot 200. FIG. 3 is a diagram illustrating an example of a hardware configuration included in the language identification server 100. FIG. 4 is a diagram illustrating an example of the functions of the service robot 200. FIG. 5 is a diagram illustrating an example of functions provided in the language identification server 100. FIG. 6 is a flowchart illustrating an example of language identification processing. FIG. 7 is a flowchart illustrating an example of language identification processing. FIG. 8 is a diagram illustrating a specific example of language identification processing. FIG. 9 is a flowchart illustrating an example of the database update process. FIG. 10 is a diagram showing an example of the database update screen. FIG. 11 is a diagram illustrating an example of tourist guidance to foreign tourists by the service robot 200.

Embodiments for carrying out the present invention will be described below with reference to the drawings.

FIG. 1 is a diagram showing an example of the configuration of a language recognition system 1 according to this embodiment. As shown in the figure, the language recognition system 1 is configured to include a language identification server 100, which is an information processing device for performing language identification, a service robot 200, and a wireless access point 300. The language recognition system 1 is a language recognition system that identifies the language of a word spoken by a tourist in a business (hereinafter referred to as the business) that mainly deals with tourists from abroad (hereinafter simply referred to as tourists). is there.

The language spoken by the tourist may be various languages such as Japanese, English, Chinese, and Korean (hereinafter referred to as target language). However, the language recognition system 1 of the present embodiment is A word (hereinafter referred to as a first language) and English (hereinafter referred to as a second language) shall be identified.

The service robot 200 is an information processing device capable of autonomous movement, which recognizes a voice uttered by a tourist and transmits the recognized voice to the language identification server 100. In addition, the service robot 200 generates a voice based on various information transmitted from the language identification server 100, and emits the generated voice to interact with the tourist.

The wireless access point 300 is an information processing device that controls communication between the service robot 200 and the language identification server 100. The wireless access point 300 is communicatively connected to the service robot 200 via the wireless communication network 5 or the like, and can communicate with the language identification server 100 via a wired communication network such as the wired LAN 4 or a wireless communication network. Connected to.

(Hardware configuration)
Next, the hardware configuration of each information processing device that constitutes the language recognition system 1 will be described.

FIG. 2 is a diagram illustrating an example of the hardware configuration of the service robot 200. As shown in the figure, the service robot 200 includes a processor 21 including a CPU (Central Processing Unit) and a main unit such as a RAM (Random Access Memory), a ROM (Read Only Memory), or an NVRAM (Non-Volatile RAM). A storage device, a storage device 22 including a storage device and an auxiliary storage device such as a hard disk (Hard Disk Drive) and an SSD (Solid State Drive), a sound collection device 23, a speaker 24, a camera 25, a mobile device 26, and wireless access. The communication device 27 that communicates with the language identification server 100 via the point 300 is provided.

The storage device 22 stores information (scenario data) about the content of the dialogue performed by the service robot 200, and a program (scenario operation program) that controls the order of the dialogue and the response method based on the scenario data.

The sound collection device 23 converts a tourist's voice into digital data. The sound collection device 23 is configured to include, for example, a microphone that converts voice into an electric signal and an AD converter that converts the electric signal output from the microphone into digital data. In the present embodiment, the sound collecting device 23 includes a plurality of microphones having different directivities, and sounds from a plurality of directions are input by these microphones.

The speaker 24 outputs sound. The speaker 24 is, for example, a DA converter that converts information (digital data) related to voice received from the language identification server 100 into an analog electric signal, an amplifier that amplifies the converted analog electric signal, and an amplified signal. And a speaker that converts an analog electric signal into physical vibration to generate sound.

The camera 25 is a photographing device that photographs a subject in a predetermined range, and recognizes, for example, the position or movement of a person facing the service robot 200.

The moving device 26 is a device for causing the service robot 200 to travel, and includes, for example, a power mechanism such as a motor and a steering mechanism such as wheels.

FIG. 3 is a diagram illustrating an example of a hardware configuration included in the language identification server 100. As shown in FIG. 1, the language identification server 100 includes a processor 11 such as a CPU (Central Processing Unit) and a RAM (Random Access Memory), a ROM (Read Only Memory), an NVRAM (Non-Volatile RAM), or the like. Via the main storage device 12, an auxiliary storage device 13 such as a hard disk and SSD (Solid State Drive), an input device 14 such as a touch panel and operation buttons, an output device 15 such as a liquid crystal display and a printer, and a wireless access point 300. And a communication device 16 for communicating with the service robot 200.

(function)
Next, the function of each information processing apparatus will be described.

FIG. 4 is a diagram illustrating an example of the functions of the service robot 200. As shown in the figure, the service robot 200 includes a voice acquisition unit 210, a voice transmission unit 220, an information reception unit 230, and a response unit 240.

The voice acquisition unit 210 acquires voice. Specifically, the voice acquisition unit 210 extracts only the human voice by removing noise and the like from the sound input to the sound collection device 23.

The voice transmitting unit 220 transmits the voice information extracted by the voice acquiring unit 210 to the language identification server 100.

The information receiving unit 230 receives information about the language (details will be described later) specified by the language identification server 100, which is transmitted from the language identification server 100.

The response unit 240 outputs the voice corresponding to the voice acquired by the voice acquisition unit 210 based on the language specified by the language identification server 100.

The functions of the service robot 200 described above are realized by the hardware of the service robot 200 or by the processor 21 of the service robot 200 reading and executing the program stored in the storage device 22.

FIG. 5 is a diagram illustrating an example of functions provided in the language identification server 100. As shown in the figure, the language identification server 100 includes a voice acquisition unit 110, a first language processing unit 120, a second language processing unit 130, a language identification unit 140, a response generation unit 150, and a database updating unit 160.

The voice acquisition unit 110 acquires voice information from the service robot 200.

The first language processing unit 120 includes a first language speech recognition unit 121, a first corresponding section calculation unit 122, and a first reliability calculation unit 123. The first language processing unit 120 also manages databases of the first language general vocabulary information 125, the first language specialized vocabulary information 126, and the first language syllable information 127.

The first-language speech recognition unit 121 includes a first-language general translation unit 1211 and a first-language specialized translation unit 1212.

The first language general translation unit 1211 is based on the first language general vocabulary information, which is information about the first vocabulary in the first language, and is the first language that is the word in the first language corresponding to the acquired voice. Generate a general translation phrase. In the present embodiment, the first vocabulary is a vocabulary that is commonly spoken in the first language and that is frequently used in the subject work (which is commonly used in the subject work or is often heard in the subject work). I shall.

The first-language specialized translation unit 1212 is a word in the first language corresponding to the acquired voice, based on first-language specialized vocabulary information that is information regarding a second vocabulary in the first language, and is a word in the first language. Generate a language-specific translation phrase. In the present embodiment, the second vocabulary is assumed to be a vocabulary that is not frequently used in the present work (it is rarely uttered in the present work or is unfamiliar with the present work). For example, the second vocabulary includes a vocabulary whose pronunciation is generally unclear by a person, such as a Japanese proper noun (a place name, etc.) issued by a foreigner visiting Japan.

The first corresponding section calculation unit 122 extracts a first language corresponding part which is a part corresponding to each other between the generated first language general translation phrase and the generated first language specialized translation phrase.

For example, the first corresponding section calculation unit extracts the first language corresponding portion based on commonality between a character string representing the first language general translation phrase and a character string representing the first language specialized translation phrase. .. Further, for example, the first corresponding section calculation unit acquires and acquires the timing of utterance of each word in the first language general translation phrase and the timing of utterance of each word in the first language specialized translation phrase. The first language corresponding portion is extracted based on the commonality of the utterance timings of the words.

The first language general vocabulary information 125 is a database (language model) that stores the first vocabulary in the first language. Specifically, the first language general vocabulary information 125 stores information such as words, fixed phrases and sentences in the first vocabulary of the first language.

The first language specialized vocabulary information 126 is a database (language model) in which the second vocabulary in the first language is stored. Specifically, the first language specialized vocabulary information 126 stores information such as words, fixed phrases, and sentences in the second vocabulary of the first language.

The first language syllable information 127 is a database that stores information (acoustic model) of voices used in the first language. The first language syllable information 127 includes, for example, information on syllables (including utterance timing and the like) in the first language.

The first reliability calculation unit 123 calculates a first reliability described later.

The second language processing unit 130 includes a second language speech recognition unit 131, a second corresponding section calculation unit 132, and a second reliability calculation unit 133. The second language processing unit 130 also manages second language general vocabulary information 135, second language specialized vocabulary information 136, and second language syllable information 137.

The second language speech recognition unit 131 includes a second language general translation unit 1311 and a second language specialized translation unit 1312.

The second language general translation unit 1311 is based on the second language general vocabulary information, which is information about the third vocabulary in the second language, and is the second language that is the word in the second language corresponding to the acquired voice. Generate a general translation phrase. Note that, in the present embodiment, the third vocabulary is a vocabulary commonly spoken in the second language and frequently used in the present work (pronounced routinely in the present work or often heard in the present work). And

The second-language specialized translation unit 1312 is a word in the second language corresponding to the acquired voice, based on the second-language specialized vocabulary information that is information about the fourth vocabulary in the second language, and the second language is the second language. Generate a language-specific translation phrase. In the present embodiment, the fourth vocabulary is assumed to be a vocabulary that is used less frequently in the present work (which is less likely to be uttered in the present work or unfamiliar with the present work). For example, the fourth vocabulary includes a vocabulary whose pronunciation is generally unclear by a person, such as a Japanese proper noun (such as a place name) issued by a foreigner who visits Japan.

The second corresponding section calculation unit 132 extracts a second language corresponding part which is a part corresponding to each other between the generated second language general translation phrase and the generated second language specialized translation phrase.

For example, the second corresponding section calculation unit 132 extracts the second language corresponding portion based on the commonality between the character string representing the second language general translation phrase and the character string representing the second language specialized translation phrase. To do. Further, for example, the second corresponding section calculation unit 132 acquires and acquires the timing of utterance of each word in the second language general translation phrase and the timing of utterance of each word in the second language specialized translation phrase. The portion corresponding to the second language is extracted based on the commonality of the utterance timings of these words.

The second language general vocabulary information 135 is a database (language model) that stores the third vocabulary in the second language. Specifically, the second language general vocabulary information 135 stores information such as words, fixed phrases, and sentences in the third vocabulary of the second language as a character string.

The second language specialized vocabulary information 136 is a database (language model) that stores the fourth vocabulary in the second language. Specifically, the second language specialized vocabulary information 136 stores, as a character string, information such as a word, a fixed phrase, and a sentence in the fourth vocabulary of the second language.

The second language syllable information 137 is a database that stores information (acoustic model) of voices used in the second language. The second language syllable information 137 includes, for example, information on syllables (including vocalization timing) in the second language.

The second reliability calculation unit 133 calculates a second reliability described later.

The language identifying unit 140 identifies the language of the input voice based on the extracted first language corresponding part and the extracted second language corresponding part.

Specifically, the language identifying unit 140 uses the first reliability that is an index indicating the certainty of the first language corresponding portion as the first language and the second reliability of the second language corresponding portion. The language of the acquired voice is specified based on the second reliability, which is an index indicating the certainty as the second language.

The reliability (first reliability, second reliability) is calculated for each word, and each reliability has a value (or score, for example, 0.0 to 1.0) within a predetermined range. The closer the value of this reliability is to the lower limit (for example, closer to 0.0), the more other words (word hypotheses) that have the same score as the word to be calculated.

Further, the reliability is obtained, for example, by calculating the posterior probability of the word hypothesis of each word in the recognized voice. Details of this calculation method are disclosed, for example, in "Confidence Measures for Large Vocabulary Continuous Speech Recognition, IEEE TRANSACTIONS ON SPEECH AND AUDIO PROCESSING, VOL. 9, NO. 3, MARCH 2001".

The response generation unit 150 outputs a voice in the specified language corresponding to the obtained voice.

The database updating unit 160 accepts an input of the first language general vocabulary information, the first language specialized vocabulary information, the second language general vocabulary information, or the second language specialized vocabulary information, and outputs the accepted information. To do.

The database updating unit 160 also includes a specialized language updating unit 161 and a specialized language conversion unit 162.

The specialized language updating unit 161 acquires the word in the first language, extracts a part of the acquired first language based on information on the frequency of use of the word, and extracts one of the extracted first languages. Part is stored as the first language specialized vocabulary information, the words of the second language are acquired, a part of the acquired second language is extracted based on information about the frequency of use of the words, and the extracted words are extracted. A part of the second language is stored as the second language specialized vocabulary information.

The specialized language conversion unit 162 converts a part of the extracted words of the first language into the words of the second language, and converts the converted words of the second language as the second language specialized vocabulary information. The part of the stored words of the second language that has been stored is converted into the words of the first language, and the converted words of the first language are stored as the first language specialized vocabulary information.

The function of the language identification server 100 described above is performed by the hardware of the language identification server 100, or the processor 11 of the language identification server 100 reads a program stored in the main storage device 12 or the auxiliary storage device 13. It is realized by executing.

Further, some of the functions of the language identification server 100 described above may be provided in the service robot 200. Further, some of the functions of the service robot 200 may be provided in the language identification server 100.

(processing)
Next, the processing performed in the language recognition system 1 will be described.

<Language identification processing>
6 and 7 are flowcharts illustrating an example of a process (hereinafter, referred to as a language identification process) of identifying a language of a word spoken by a tourist among the processes performed in the language recognition system 1 (for convenience of space). It is divided into two figures). The language identification processing is started, for example, when the language identification server 100 and the service robot 200 are powered on.

As shown in the figure, the service robot 200 first waits for reception of voice from a tourist (S1).

When the voice is received, the service robot 200 removes sounds other than the voice of the tourist (such as noise and noise) from the received voice to acquire only the voice of the tourist (S3). Specifically, the service robot 200 removes noise based on the acquired phase differences of the voices of the plurality of channels. The sound from which noise and the like have been removed is, for example, only the sound (one-channel sound) acquired by the service robot 200 from the front direction.

The service robot 200 transmits the voice acquired in S3 to the language identification server 100 (S5).

When the language identification server 100 receives a voice from the service robot 200 (S7), based on the first language general vocabulary information 125, the character string of the word of the first language (the first language corresponding to the voice received in S7) A general translation phrase is generated (S9).

In addition, the language identification server 100 generates a character string of a word in the first language (first language specialized translation phrase) corresponding to the voice received in S7, based on the first language specialized vocabulary information 126 (S11). ..

Then, the language identification server 100 extracts a portion (first language corresponding portion) that mutually corresponds between the first language general translation phrase generated in S9 and the first language specialized translation phrase generated in S11 ( S13). Specifically, the language identification server 100 extracts, as the first language corresponding part, a common part of the character string corresponding to the first language general translation phrase and the character string corresponding to the first language specialized translation phrase.

Then, the language identification server 100 calculates the first reliability of the first language extraction extracted in S13 (S15).

In addition, the language identification server 100 generates a character string of a word in the second language (second language general translation phrase) corresponding to the voice received in S7, based on the second language general vocabulary information (S17). In addition, the language identification server 100 generates a character string of a word in the second language (second language specialized translation phrase) corresponding to the voice received in S7 based on the second language specialized vocabulary information (S19).

Further, the language identification server 100 extracts a portion (second language corresponding portion) in which the character string of the second language generated in S17 and the character string of the second language generated in S19 correspond to each other. Yes (S21). Specifically, the common part of the character string corresponding to the second language general translation phrase and the character string corresponding to the second language specialized translation phrase is extracted as the second language corresponding part.

Then, the language identification server 100 calculates the second reliability of the second language corresponding part extracted in S21 (S23).

The language identification server 100 identifies the language of the voice received in S7 based on the first reliability calculated in S15 and the second reliability calculated in S23 (S25). Specifically, the language identification server 100 compares the first reliability and the second reliability, and stores the language with the higher reliability as the language of the voice received in S7.

Subsequently, the language identification server 100 analyzes the meaning content of the voice received in S7 (S27).

Specifically, the language identification server 100 acquires the character string of the language specified in S25, which corresponds to the voice received in S7. For example, when the language specified in S25 is the first language, the language identification server 100 acquires the character string generated in S9 or S11, and when the language specified in S25 is the second language, The language identification server 100 acquires the character string generated in S17 or S19. Then, the language identification server 100 analyzes the meaning content of the acquired character string.

The language identification server 100 transmits the content analyzed in S27 to the service robot 200 (S29).

The service robot 200 generates a voice based on the information received from the language identification server 100 (S31, S33). Specifically, for example, the service robot 200 converts the information of the dialogue indicated by the scenario data into voice according to the received information and the scenario operation program.

Then, the service robot 200 outputs the voice generated in S33 (S35). With that, the language identification process ends (S37).

<Specific example of language identification processing>
Here, a specific example of the language identification processing will be described.
FIG. 8 is a diagram illustrating a specific example of language identification processing. As shown in the figure, first, it is assumed that the language identification server 100 receives voice information "How to get to KASAI RINKAI KOEN?" (hereinafter referred to as the voice) from the service robot 200 (reference numeral 1001).

Then, the language identification server 100 assumes that the language of this voice is the first language (Japanese) based on the first language specialized vocabulary information 126, and calls this voice "Hatsutsuki Kasai Rinkai Park". Convert to a character string (reference numeral 1002). The language identification server 100 recognizes this character string as a combination of the character string “first month” and the character string “Kasai Rinkai Park”.

Further, the language identification server 100 assumes that the language of this voice is the first language (Japanese) based on the first language general vocabulary information 125, and calls this voice "Hatsutsuki Fire Seaside Park". Convert to a character string (reference numeral 1003). The language identification server 100 recognizes this character string as a combination of the character string “first month” and the character string “fire coastal park”.

The language identification server 100 extracts the first language corresponding part based on the character string shown by reference numeral 1002 and the character string shown by reference numeral 1003 (reference numeral 1004). That is, the language identification server 100 extracts “first month”, which is a continuous common character string between the character string shown by the reference numeral 1002 and the character string shown by the reference numeral 1003, as the first language corresponding part. To do.

When there are a plurality of common character strings, the common character string having the maximum length is extracted from the common character strings. The same applies to the second language corresponding part described later.

Next, the language identification server 100 assumes that the language of the main voice is the second language (English) based on the second language specialized vocabulary information 136, and outputs the main voice as "How to get to KASAI RINKAI KOEN. It is converted into a character string "?" (reference numeral 1005). The language identification server 100 recognizes this character string as a combination of the character string “How to get to” and the character string “KASAI RINKAI KOEN”.

In addition, the language identification server 100 assumes that the language of the main voice is the second language (English) based on the second language general vocabulary information 135, and outputs the main voice as "How to get to as I bring eye. It is converted into a character string "hole?" (reference numeral 1006). The language identification server 100 recognizes this character string as a combination of the character string “How to get to” and the character string “as I bring eye hole”.

The language identification server 100 extracts the second language corresponding part based on the character string shown by reference numeral 1005 and the character string shown by reference numeral 1006 (reference numeral 1007). That is, the language identification server 100 determines that the continuous common character string “How to get to” between the character string 1005 and the character string 1006 is the second language corresponding part. To extract.

The language identification server 100 may, instead of or in addition to extracting the second language corresponding part based on the character string as described above, utter the timing of uttering each word in the first language general translation phrase, and The second language corresponding part may be extracted based on the timing of utterance of each word in the language-specific translation phrase (based on the acoustic model). That is, the language identification server 100 determines "How to get to" between the utterance of the character string 1005 and the utterance of the character string 1006.
Since the timing at which each word is uttered and the timing at which there is no sound are common in the portion, "How to get to" is extracted as the second language corresponding portion. It should be noted that such a method may be similarly adopted for the portion corresponding to the first language.

Then, the language identification server 100 sets the reliability (first reliability) of the first language corresponding part indicated by reference numeral 1004 and the reliability (second reliability) of the second language corresponding part indicated by reference numeral 1007. It is calculated (reference numeral 1008).

For example, when the language identification server 100 determines that the first reliability is higher than the second reliability, the first language (Japanese), which is the language of the first language corresponding part for which the first reliability is calculated. Is recognized as a language corresponding to the voice indicated by reference numeral 1001.

<Database update process>
Next, the update processing of each database performed in the language recognition system 1 will be described.

FIG. 9 is a flowchart illustrating an example of a process of updating a database (hereinafter referred to as a database update process). The database update process is started, for example, when a predetermined operation input is performed on the language identification server 100.

As shown in the figure, first, the language identification server 100 receives an input of a word in the first language (for example, a sentence or a word; hereinafter referred to as business data) used in the business of this case from the user (S41). .. Note that the business data may be the first language generated based on the voice acquired by the language identification server 100 (voice received by the service robot 200).

When the input of the business data is accepted, the language identification server 100 extracts the second vocabulary of the first language (technical terms, proper nouns and the like originated by foreigners) from the business data and extracts the second vocabulary. The vocabulary is stored in the first language specialized vocabulary information 126 (S42). The second vocabulary is extracted, for example, based on information stored in the language identification server 100 about the usage frequency of each word in the first language. For example, the language identification server 100 stores the business data in the first language specialized vocabulary information 126 when the frequency of use of the word indicated by the business data is less than or equal to a predetermined value or within a predetermined range.

In addition, the language identification server 100 converts the second vocabulary extracted in S42 into a fourth vocabulary, and stores the converted fourth vocabulary in the second language specialized vocabulary information 136 (S43). With that, the process ends (S44).

The database update process described above is performed not only for the first language but also for the second language. That is, the language identification server 100 extracts the fourth vocabulary of the second language (technical terms, proper nouns issued by foreigners, etc.) from the business data, and extracts the fourth vocabulary as the second language specialized vocabulary. Information 136 is stored. In addition, the language identification server 100 converts the extracted fourth vocabulary into the second vocabulary, and stores the converted second vocabulary in the first language specialized vocabulary information 126.

<Database update screen>
Here, the screen displayed in the database update process will be described.

FIG. 10 is a diagram showing an example of a screen for editing each database (hereinafter referred to as a database update screen) displayed in the database update process. As shown in the figure, the database update screen includes a general language model edit screen 500 and a specialized language model edit screen 600.

The general language model edit screen 500 receives input of the first vocabulary or the third vocabulary. The general language model edit screen 500 includes a first language input field 501, a second language input field 502, and a registration confirmation field 503. The first language input field 501 accepts input of a character string (first vocabulary) in the first language. In the second language input field 502, input of a character string in the second language (third vocabulary) is accepted. In the registration confirmation field 503, the character string input in the first language input field 501 is stored in the first language general vocabulary information 125, and the character string input in the second language input field 502 is stored in the second language general vocabulary information. An input to the effect that it is stored in 135 is accepted.

The specialized language model edit screen 600 receives input of the second vocabulary or the fourth vocabulary. The specialized language model edit screen 600 includes a first language input field 601, a second language input field 602, and a registration confirmation field 603. The first language input field 601 accepts input of a character string (second vocabulary) in the first language. The second language input field 602 accepts input of a character string in the second language (fourth vocabulary). In the registration confirmation field 603, the character string input in the first language input field 601 is stored in the first language specialized vocabulary information 126, and the character string input in the second language input field 602 is stored in the second language specialized vocabulary information. An input to store the data in 136 is accepted.

<Example of tourist information by service robot>
Here, an example of sightseeing guidance by the service robot 200 will be described.
FIG. 11 is a diagram illustrating an example of tourist guidance to foreign tourists by the service robot 200. As shown in the figure, the service robot 200 equipped with a 14-channel microphone has one of a plurality of languages (Japanese, English, Chinese, Korean) issued by a foreign tourist ( Hereinafter, a word in an input language is input (reference numeral 1101). Other unnecessary voice such as noise that is input at the same time as the voice of this word is removed by a predetermined noise suppression process, and becomes one-channel voice (hereinafter referred to as input voice).

Upon receiving the input voice from the service robot 200, the language identification server 100 performs language identification processing.

Specifically, the language identification server 100 determines the reliability (first reliability) of the input voice as Japanese (first reliability), first language general vocabulary information 125, first language specialized vocabulary information 126, and first language syllables. It is calculated based on the information 127 (reference numeral 1102). Also, the language identification server 100 determines the reliability (second reliability) of the input voice as English based on the second language general vocabulary information 135, the second language specialized vocabulary information 136, and the second language syllable information 137. It is calculated (reference numeral 1103). Although not shown, the language identification server 100 similarly calculates the reliability of each language for other languages (Chinese, Korean, etc.).

Then, the language identification server 100 compares the calculated reliabilities and specifies that the language with the highest reliance (English in this example) is the input language of the input voice (reference numeral 1104). Further, the language identification server 100 recognizes that the input voice is a question of how to reach a predetermined destination by analyzing the semantic content of the specified language (the semantic content of the input voice), and displays the recognized result. It is transmitted to the service robot 200 (reference numeral 1105).

Based on the received information, the service robot 200 utters a voice corresponding to the above question, that is, utters a sentence explaining how to reach a predetermined destination (reference numeral 1106).

As described above, according to the language recognition system 1 of the present embodiment, the first-language general translation phrase generated based on the first-language general vocabulary information and the first-language specialized translation generated based on the first-language specialized vocabulary information. A second language general translation phrase generated based on the second language general vocabulary information and a second language specialty generated based on the second language specialized vocabulary information by extracting the first language corresponding parts corresponding to the phrases. The second language corresponding part corresponding to the translation phrase is extracted, and the language of the voice is specified based on the first language corresponding part and the second language corresponding part. That is, for the first language and the second language, the language is recognized based on the corresponding portions between the words extracted based on the information of the different vocabulary, and therefore only the portion suitable for performing the language recognition is extracted. Can be identified.

For example, the language can be identified based on the part where the speech recognition result by the language model of the specialized vocabulary corresponds to the speech recognition result by the language model of the general vocabulary. Therefore, even if there is a risk that an error may occur in the voice recognition result due to the inclusion of foreign words in the words spoken by the speaker, the language spoken by the speaker can be accurately recognized.

As described above, according to the language recognition system 1 of the present embodiment, it is possible to accurately perform language recognition.

Further, according to the language recognition system 1 of the present embodiment, the first language corresponding portion is extracted based on the commonality between the character string representing the first language general translation phrase and the character string representing the first language specialized translation phrase. , The second language corresponding part is extracted based on the commonality between the character string representing the second language general translation phrase and the character string representing the second language specialized translation phrase, so that the first language corresponding part and the second language corresponding part are extracted. In, it is possible to reliably extract a part having a common meaning. Thereby, language recognition can be accurately performed.

Further, according to the language recognition system 1 of the present embodiment, the first language is based on the commonality of the utterance timing of each word in the first language general translation phrase and the utterance timing of each word in the first language specialized translation phrase. Since the corresponding portion is extracted and the second language corresponding portion is extracted based on the commonality of the utterance timing of each word in the second language general translated phrase and the utterance timing of each word in the second language specialized translated phrase, Language recognition can be performed based on the timing and rhythm of vocalization that is unique to the language. Thereby, language recognition can be accurately performed.

Further, according to the language recognition system 1 of the present embodiment, the first reliability indicating the certainty of the first language corresponding part as the first language and the second reliability of the second language corresponding part as the second language. Since the language of the voice is specified based on the second reliability indicating the certainty, it is possible to perform the language recognition with stable accuracy.

Further, according to the language recognition system 1 of the present embodiment, the input of the first language general vocabulary information, the first language specialized vocabulary information, the second language general vocabulary information, or the second language specialized vocabulary information is accepted, and the input is accepted. Since the information is output, the user of the language recognition system 1 can appropriately update each information while checking the input content by himself/herself to improve the accuracy of language recognition.

Further, according to the language recognition system 1 of the present embodiment, since the voice of the specified language corresponding to the uttered voice is output, an accurate dialogue with a person is possible. For example, it is possible to appropriately perform automatic voice dialogue with a foreign tourist by a robot.

Further, according to the language recognition system 1 of the present embodiment, a part of the acquired first and second languages is extracted based on the information on the frequency of use of words, and a part of the extracted first and second languages is extracted. Is stored as the first and second language specialized vocabulary information, the first and second language specialized vocabulary information can be updated to an appropriate content. As a result, the accuracy of language recognition can be improved.

Further, according to the language recognition system 1 of the present embodiment, a part of the extracted words of the first and second languages is converted into the words of the second and first languages, and the converted second and first words are converted. Since the words of the language are stored as the second and first language specialized vocabulary information, the vocabularies of the first language and the second language can be stored at the same time. As a result, the accuracy of language recognition can be improved.

The above description of the embodiments is for facilitating the understanding of the present invention and does not limit the present invention. The present invention can be modified and improved without departing from the spirit thereof and the present invention includes equivalents thereof.

For example, in the present embodiment, the case has been described in which the language recognition system 1 is a system that recognizes words spoken by tourists, but the speaker is not limited to tourists, and may be any person in general. For example, the service robot 200 may be a nursing robot or a nursing robot, and the speaker may be a patient or a doctor in a hospital facility or a nursing facility.

Further, the language identification server 100 may extract the first language corresponding part and the second language corresponding part from previously stored vocabulary information relating to the first language corresponding part and the second language corresponding part.

Further, in the present embodiment, only the first language (Japanese) and the second language (English) are given as the target languages, but the type and number of languages are not limited to this, and Chinese, Other languages such as Korean can also be the target language for language recognition.

1 language recognition system, 100 language identification server, 200 service robot, 210 voice acquisition unit, 1211 first language general translation unit, 1212 first language specialized translation unit, 122 first corresponding section calculation unit, 1311 second language general translation unit , 1312 second language specialized translation unit, 132 second corresponding section calculation unit, 133 second reliability calculation unit, 140 language identification unit, 150
Response generator

Claims (15)

A language recognition system, comprising a processor and a memory, for recognizing a language of an emitted voice,
A voice acquisition unit that acquires voice,
A first language that generates a first language general translation phrase that is a word of the first language corresponding to the acquired voice, based on first language general vocabulary information that is information about the first vocabulary in the first language. General translation department,
A first language specialized translation phrase, which is a word in the first language corresponding to the acquired voice, is generated based on first language specialized vocabulary information that is information about a second vocabulary in the first language. A specialized translation department,
A first corresponding section calculation unit that extracts a first language corresponding portion that is a mutually corresponding portion between the generated first language general translation phrase and the generated first language specialized translation phrase;
A second language that generates a second language general translation phrase that is a word of the second language corresponding to the acquired voice, based on second language general vocabulary information that is information about the third vocabulary in the second language. General translation department,
A second language specialized translation phrase, which is a word in the second language corresponding to the acquired voice, is generated based on second language specialized vocabulary information that is information about a fourth vocabulary in the second language. A specialized translation department,
A second corresponding section calculation unit that extracts a second language corresponding portion that is a mutually corresponding portion between the generated second language general translation phrase and the generated second language specialized translation phrase;
A language identification unit that identifies a language of the acquired voice based on the extracted first language corresponding part and the extracted second language corresponding part;
A language recognition system equipped with. The first corresponding section calculation unit extracts the first language corresponding portion based on commonality between a character string representing the first language general translation phrase and a character string representing the first language specialized translation phrase,
The second corresponding section calculation unit extracts the second language corresponding portion based on the commonality between the character string representing the second language general translation phrase and the character string representing the second language specialized translation phrase.
The language recognition system according to claim 1. The first corresponding section calculation unit acquires the timing of utterance of each word in the first language general translation phrase and the timing of utterance of each word in the first language specialized translation phrase, and the acquisition of these acquired words Based on the timing commonality, the first language corresponding part is extracted,
The second corresponding section calculation unit acquires the timing of utterance of each word in the second language general translation phrase and the timing of utterance of each word in the second language specialized translation phrase, and the acquired utterance of these words. The second language corresponding part is extracted based on the commonality of the timings of
The language recognition system according to claim 1. The language identification unit has a first reliability, which is an index indicating the certainty of the first language corresponding portion as the first language, and an authenticity of the second language corresponding portion as the second language. Specify the language of the acquired voice based on the second reliability, which is an index indicating the likelihood,
The language recognition system according to claim 1. The language recognition system according to claim 1, further comprising: a response unit that outputs a voice in the specified language corresponding to the acquired voice. A database that accepts an input of at least one of the first language general vocabulary information, the first language specialized vocabulary information, the second language general vocabulary information, and the second language specialized vocabulary information, and outputs the accepted information. The language recognition system according to claim 1, further comprising an updating unit. Gets the words of the first language, some words of the acquired first language, and extracted based on the information about the frequency of use of words, the part of the words of the extracted first language Stored as vocabulary information specialized in the first language,
The second retrieves the words of the language, some of the words of the acquired second language, and extracted based on the information about the frequency of use of words, extracted the part of the words of the second language Store as second language specialized vocabulary information,
The language recognition system according to claim 1, further comprising a specialized language updating unit. Converting a part of the extracted words of the first language to words of the second language, storing the converted words of the second language as the second language specialized vocabulary information,
A part of the extracted words of the second language is converted into the words of the first language, and the converted words of the first language are stored as the first language specialized vocabulary information;
The language recognition system according to claim 7, further comprising a specialized language conversion unit. A response unit that outputs a voice in the specified language corresponding to the obtained voice,
A database that accepts an input of at least one of the first language general vocabulary information, the first language specialized vocabulary information, the second language general vocabulary information, and the second language specialized vocabulary information, and outputs the accepted information. Update department,
Gets the words of the first language, some words of the acquired first language, and extracted based on the information about the frequency of use of words, the part of the words of the extracted first language Stored as vocabulary information specialized in the first language,
The second retrieves the words of the language, some of the words of the acquired second language, and extracted based on the information about the frequency of use of words, extracted the part of the words of the second language A specialized language update unit that stores the second language specialized vocabulary information;
Converting a part of the extracted words of the first language to words of the second language, storing the converted words of the second language as the second language specialized vocabulary information,
A specialized language conversion unit that converts a part of the extracted words of the second language into words of the first language and stores the converted words of the first language as the first language specialized vocabulary information; Equipped with
The first corresponding section calculation unit extracts the first language corresponding portion based on commonality between a character string representing the first language general translation phrase and a character string representing the first language specialized translation phrase, and / Or, the timing of utterance of each word in the first language general translation phrase and the timing of utterance of each word in the first language specialized translation phrase are acquired, and the acquired timing of utterance of these words is common. Based on the above, the first language corresponding part is extracted,
The second corresponding section calculation unit extracts the second language corresponding portion based on commonality between a character string representing the second language general translation phrase and a character string representing the second language specialized translation phrase, and Or, the timing of utterance of each word in the second language general translation phrase and the timing of utterance of each word in the second language specialized translation phrase are acquired, and commonality of the acquired utterance timings of these words is acquired. Based on the second language corresponding portion is extracted,
The language identification unit has a first reliability, which is an index indicating the certainty of the first language corresponding portion as the first language, and an authenticity of the second language corresponding portion as the second language. Specify the language of the acquired voice based on the second reliability, which is an index indicating the likelihood,
The language recognition system according to claim 1. A language recognition method for recognizing the language of a spoken voice,
An information processing device including a processor and a memory,
A voice acquisition process that acquires voice,
A first language that generates a first language general translation phrase that is a word of the first language corresponding to the acquired voice, based on first language general vocabulary information that is information about the first vocabulary in the first language. General translation processing,
A first language specialized translation phrase, which is a word in the first language corresponding to the acquired voice, is generated based on first language specialized vocabulary information that is information about a second vocabulary in the first language. Language specialized translation processing,
A first corresponding section calculation process for extracting a first language corresponding portion which is a mutually corresponding portion between the generated first language general translation phrase and the generated first language specialized translation phrase;
A second language that generates a second language general translation phrase that is a word of the second language corresponding to the acquired voice, based on second language general vocabulary information that is information about the third vocabulary in the second language. General translation processing,
A second language specialized translation phrase, which is a word in the second language corresponding to the acquired voice, is generated based on second language specialized vocabulary information that is information about a fourth vocabulary in the second language. Language specialized translation processing,
A second corresponding section calculation process for extracting a second language corresponding portion which is a mutually corresponding portion between the generated second language general translation phrase and the generated second language specialized translation phrase;
A language identification process for specifying the language of the acquired voice based on the extracted first language corresponding part and the extracted second language corresponding part;
A language recognition method that performs. In the first corresponding section calculation process, the first language corresponding portion is extracted based on the commonality between the character string representing the first language general translation phrase and the character string representing the first language specialized translation phrase,
The second corresponding section calculation process includes a process of extracting the second language corresponding part based on the commonality between a character string representing the second language general translation phrase and a character string representing the second language specialized translation phrase. Including,
The language recognition method according to claim 10. Gets the words of the first language, some words of the acquired first language, and extracted based on the information about the frequency of use of words, the part of the words of the extracted first language Stored as vocabulary information specialized in the first language,
The second retrieves the words of the language, some of the words of the acquired second language, and extracted based on the information about the frequency of use of words, extracted the part of the words of the second language Store as second language specialized vocabulary information,
The language recognition method according to claim 10, wherein a specialized language update process is executed. A language recognition program for recognizing the language of voices emitted,
An information processing device including a processor and a memory,
A voice acquisition process that acquires voice,
A first language that generates a first language general translation phrase that is a word of the first language corresponding to the acquired voice, based on first language general vocabulary information that is information about the first vocabulary in the first language. General translation processing,
A first language specialized translation phrase, which is a word in the first language corresponding to the acquired voice, is generated based on first language specialized vocabulary information that is information about a second vocabulary in the first language. Language specialized translation processing,
A first corresponding section calculation process for extracting a first language corresponding portion which is a mutually corresponding portion between the generated first language general translation phrase and the generated first language specialized translation phrase;
A second language that generates a second language general translation phrase that is a word of the second language corresponding to the acquired voice, based on second language general vocabulary information that is information about the third vocabulary in the second language. General translation processing,
A second language specialized translation phrase, which is a word in the second language corresponding to the acquired voice, is generated based on second language specialized vocabulary information that is information about a fourth vocabulary in the second language. Language specialized translation processing,
A second corresponding section calculation process for extracting a second language corresponding portion which is a mutually corresponding portion between the generated second language general translation phrase and the generated second language specialized translation phrase;
A language identification process for specifying the language of the acquired voice based on the extracted first language corresponding part and the extracted second language corresponding part;
A language recognition program that executes. In the first corresponding section calculation process, the first language corresponding portion is extracted based on the commonality between the character string representing the first language general translation phrase and the character string representing the first language specialized translation phrase,
The second corresponding section calculation process includes a process of extracting the second language corresponding part based on the commonality between a character string representing the second language general translation phrase and a character string representing the second language specialized translation phrase. Including,
The language recognition program according to claim 13. Gets the words of the first language, some words of the acquired first language, and extracted based on the information about the frequency of use of words, the part of the words of the extracted first language Stored as vocabulary information specialized in the first language,
The second retrieves the words of the language, some of the words of the acquired second language, and extracted based on the information about the frequency of use of words, extracted the part of the words of the second language Store as second language specialized vocabulary information,
The language recognition program according to claim 13, which executes a specialized language update process.