JP4067483B2 - Telephone reception translation system - Google Patents

Description

  The present invention identifies a call partner or a language spoken by a call partner from a received phone call pattern, etc., and based on the specified language type, accepts a call by recognizing the content spoken by the call partner and translating it into a desired language. The present invention relates to a translation system, and more particularly to a telephone reception translation system suitable for application to a call center or the like that receives a telephone call from a specific customer.

As a system for returning an answer to a question from a customer to the customer, various customer response systems using an automatic voice response system have been proposed (see, for example, Patent Document 1). The one described in Patent Document 1 introduces voice recognition so that information obtained by voice can be sent back immediately, and the customer who has accessed is specified, and the customer attribute is set using the customer name as a key. Various services are realized by searching a database storing information, selecting a message suitable for the customer, and making an automatic voice response.
On the other hand, a call center or the like sometimes receives a call from a customer who speaks a language other than Japanese or a special dialect even in the case of Japanese (hereinafter referred to as a language other than Japanese). In this case, it is necessary to transfer the call to a person who can speak the language, but it is difficult to secure the necessary number of persons who can speak various languages, and it is not possible to deal with it sufficiently. .

As described above, various systems for providing a telephone translation service have been proposed in order to facilitate a call with a speaker other than Japanese.
For example, in Patent Document 2, referring to the language and call history of a call partner stored in advance, the use of a translated telephone service is automatically set, and voice recognition is performed by a voice recognition device built in the telephone device. Thus, a translation telephone device is described in which character information as a speech recognition result is supplied to a translation telephone device for translation.
JP 2003-169147 A JP 2002-118659 A

In a system that translates a language, if the language used cannot be identified, translation cannot be performed, and identification of the language used is essential.
In a conventional system that translates a language for a received phone call, it is common to identify a region from the notified telephone number, determine the language type spoken in that region, and translate the language. In addition, there is a system that specifies and translates the language type spoken in the area based on the position information of GPS where the other party is located.
The method of determining the language type from the notified phone number is that the phone number alone is used only when the phone is used by multiple people who speak various languages. Cannot be determined. In addition, when the other party makes a phone call from a place other than the place where he / she usually makes a call due to a business trip or the like (or in another country), the language type may be specified incorrectly and the translation may not be performed correctly. Also, the identification of the language type based on the telephone number is impossible because the telephone number is unknown when the other party is in the telephone number non-notification mode.
Furthermore, the method of using the GPS position information must transmit GPS information, such as using a telephone capable of transmitting GPS position information. In a normal telephone, the position of the other party is specified from the GPS information. Is difficult.
The present invention has been made to solve the above-mentioned problems of the prior art, and is a system capable of specifying the language spoken by the call partner and translating the language regardless of where the call partner calls. The purpose is to provide.

In the present invention, the above problem is solved as follows.
In the call acceptance translation system that receives a call from the other party, translates the language spoken by the other party, and talks to the other party, the first voice produced by the other party and the probability of the appearance of that voice And the voice pattern storage means that stores the number of appearances corresponding to the personal name, the first voice pattern that the other party calls when receiving a call, and the language type registered in advance, A speech pattern for each language type that is considered to respond to the speech, a language type storage means that stores the appearance probability of the appearance of the speech pattern, an initial speech pattern that the other party utters when receiving a call, and the speech pattern storage collates the voice pattern registered in the means the appearance order of probability, and personal identification means for identifying the other party, person is identified by the personal identification means In addition, referring to the language type information storing the language type corresponding to the specified personal name, the language type extraction means for specifying the language of the other party and the language spoken by the other party are translated according to the specified language type. When the person is specified by the individual specifying means, the language type information storing the language type corresponding to the specified personal name is referred to specify the language of the other party and specified. The language spoken by the other party is translated according to the selected language type. In addition, when the other party cannot be identified, the first voice pattern emitted by the other party is compared with the voice pattern stored in the language type storage means in the order of appearance probability, the other party's language is identified, and the identified call The language type information spoken by the other party is registered in the voice pattern storage means in association with the personal name.

In the present invention, the following effects can be obtained.
(1) Since the type of language is determined by identifying the other party, the language spoken by the other party can be translated no matter where the other party calls.
(2) Even if the language type spoken by the call partner is unknown for the first call partner, the language type of the call partner can be specified by means of specifying the language type and can be translated.
Further, when the same person calls, it is possible to specify the individual from the second time and immediately specify the language type.
Furthermore, if a person is identified and a voice recognition process is performed using a voice recognition dictionary prepared for each person, the voice recognition accuracy can be improved.

FIG. 1 is a diagram showing a system configuration of an embodiment of the present invention.
In the figure, reference numeral 100 denotes an exchange, and the exchange 100 is connected via a public network 101. Each exchange 100 is connected to a telephone 103b of a call partner, a telephone 103a and a processing device 102 provided on the call center side, for example. Is connected.
The processing device 102 receives a call from the other party, specifies the other party's language type by specifying the other party, as will be described later, and performs speech recognition to perform translation processing.
In FIG. 1, when the processing device 102 is provided only on the telephone 103 a side, and the language type of the other party is, for example, English, the content of the other party speaking sent via the public network 101 is displayed in the processing unit 102 in Japan. This shows a case where it is translated into a word or the like and transmitted to the correspondent's telephone 103a by voice or the like, and the Japanese spoken by the correspondent is translated into English or the like by the processing device 102 and transmitted to the other party's telephone via the public network. In the following embodiment, this case will be described. However, for example, the following configuration may be used.
(1) A processing device is also provided on the telephone 103b side of the call partner, and the processing device translates English or the like spoken by the call partner into Japanese and transmits it to the correspondent phone 103a via the public network 101. The Japanese spoken by the person is translated into English by the processing unit 102 on the correspondent side and transmitted to the telephone of the other party via the public network.
(2) A processing device is also provided on the telephone 103b side of the call partner, and the English or the like spoken by the call partner is sent as it is to the processing device 102 via the public network 101, and the processing device 102 translates it into Japanese etc. Is communicated to the telephone 103a of the responder, the Japanese spoken by the correspondent is transmitted to the other party via the public network 101 as it is, and the Japanese is translated into English by the processing unit on the other party and transmitted to the other party. .

FIG. 2 is a block diagram illustrating a functional configuration of the processing apparatus according to the first embodiment of this invention.
The processing device 102 includes a voice pattern registration means 10 for registering the voice of the call partner A in the voice pattern storage unit 11, a voice pattern of the first voice of the call partner A registered in the voice pattern storage unit 11, and a call Personal identification means 12 is provided for collating the voice pattern stored in the voice pattern table 13 for each party and identifying the name of the other party.
When the individual name of the other party is specified by the individual specifying means 12, the language specifying means 14 refers to the personal name and the language type information table 17 in which the language type is registered corresponding to the personal name, and the call is made. The language type of the other party is specified, and the specified language type is registered in the language type information registration unit 15.
If the personal name of the calling party A cannot be specified, the language specifying unit 14 compares the voice pattern of the first voice of the calling party with the voice pattern stored in the language-specific voice pattern table 16, and The language type is specified, and the specified language type is registered in the language type information registration unit 15.

When the language type of the communication partner A is specified, the voice recognition means 18 refers to the language-specific voice recognition dictionary 19 and uses the recognition dictionary corresponding to the language type registered in the language type information registration unit 15. The call contents of the call partner registered in the voice pattern storage unit 11 are converted into character information.
In addition, the translation means 20 refers to the language-specific translation dictionary 22 and uses the recognition dictionary corresponding to the language type registered in the language type information registration unit 15 so that the call counterpart B can call Translate to a spoken language (explained as Japanese below).
The content of the call translated into Japanese is output as voice from the display / voice output means 21 or as character information on the display device of the processing device 102.
When the responder B responds to the caller A with Japanese voice or character input according to the translation result, the response content is converted into character information by voice recognition means in the case of voice, In the case of character input, it is sent to the translation means 20 as it is, translated into the language of the other party A, and sent to the other party A via the public network 101 described above.
If the language identification unit 14 does not identify the language of the call partner, the transfer unit 23 transfers the received call to the correspondent C who can handle multiple languages.

FIG. 3 is a flowchart showing the processing of the present embodiment. The processing of the present embodiment will be described with reference to FIG.
In FIG. 3, when the other party's call is received, the other party's voice is captured and stored in the voice pattern storage unit 11 (step S1). Further, the individual specifying means 12 compares the first voice of the call partner with the voice pattern stored in the call partner-specific voice pattern table, and specifies the call partner (step S2). In the above-mentioned voice pattern table for each call partner, the personal name of the call partner, the voice pattern of the first voice of the call partner and the appearance probability thereof are registered. The voice pattern registered in this table and the call partner's voice pattern are registered. By comparing the first voice, the other party is specified (specific details of the other party will be described later).
When the personal name of the other party is specified, the language type information table 17 is referred to, the language type of the other party is specified, and the specified language type is registered in the language type information registration unit 15. FIG. 4 is a diagram illustrating an example of the language type information table 17. In the language type information table, a personal name and a language type spoken by the person are registered in correspondence with the personal name as shown in the figure, and the language type is specified by specifying the personal name.

On the other hand, if the personal name is not specified, the language specifying unit 14 compares the voice pattern of the first voice of the other party with the voice pattern stored in the language-specific voice pattern table 16 to determine the language spoken by the other party. The type is specified (steps S9 and S10).
The language-specific voice pattern table 16 registers the language type, the voice pattern of the first voice of the other party for each language type, and the appearance probability thereof. The voice pattern registered in this table and the other party's voice pattern are registered. The language type is specified by comparing the first voice. When the language type is specified, the specified language type is registered in the language type information registration unit 15 (details of specifying the language type will be described later).
If the language type is not specified, the call is transferred to the correspondent C who can handle the call (steps S11 and S12).
When the language type is specified, the speech recognition unit 18 refers to the language-specific speech recognition dictionary 19 to convert it into character information (step S4), and the language translation dictionary 22 refers to the translation unit 20 to translate it. (Step S5).
If the voice recognition means 18 or the translation means 20 cannot translate, for example, the voice of the other party is not clear, as described above, the call is transferred to the correspondent C who can handle the call (steps S6 and S12). If translation is possible, the translation result is output as audio or as character information on a display device.
The processes in steps S4 to S7 are repeated for the call contents of the call partner A and the correspondent B, and the process ends when the call ends.
In the above embodiment, a language-specific recognition dictionary that is commonly used for the other party is used as the voice recognition dictionary. In addition, a voice recognition dictionary is prepared for each other party. Voice recognition may be performed using the recognition dictionary. Thereby, the voice recognition accuracy can be improved.

Next, the personal name identification process of the call partner by the personal identification means will be described in detail.
FIG. 5 is a diagram showing an example of the voice pattern table 13 for each call partner. As shown in FIG. 5, in the voice pattern table for each caller, a personal name, a voice pattern, an appearance probability of appearance of the voice pattern, and the number of appearances are registered in association with the personal name.
For example, in the case of Japanese, it is assumed that the other party speaks the following voice with the first voice.
(A) "Moshimoshi."
(B) “Hi, Yamadades.”
(C) “Yamadades.”
In this embodiment, in the above response, a voice pattern is registered in units of one syllable consisting of a group of voices (for example, “Moshimoshi”, “High”, “Yamadades”, etc.) The voice pattern of one syllable is compared with the voice pattern.
It is considered that the other party responds with the above (a), (b), or (c) depending on the time and circumstances, and the appearance probability indicates the probability that the first voice pattern appears as described above. is there.

FIG. 6 is a flowchart showing the individual identification process.
First, the first voice of the calling party is captured (step S1), and the first voice pattern A is compared with the voice pattern in the calling party voice pattern table. That is, the voice pattern B is extracted from the voice pattern by call partner in descending order of appearance rate, the voice patterns A and B are compared, and the hit rate is stored in the pattern hit rate registration means (steps S2 and S3).
FIG. 7 shows a registration example of the pattern hit rate registration means. The pattern hit rate registration means stores a hit rate which is a comparison result between the voice pattern A and the voice pattern B extracted from the caller-specific voice pattern table in descending order of appearance rate.
When the comparison is completed, the pattern hit rate registered in the pattern hit rate registering unit is read out, and a search is made for a hit rate exceeding a certain value (steps S4 and S5).
If there is a hit rate exceeding a certain value, the one with a high hit rate is identified as the call partner (steps S6 and S7). If there is no item with a high hit rate, it is determined that the individual cannot be identified, and the process is terminated (step S8). For example, in the example shown in FIG. 7, since the hit rate with the voice pattern B1 of the personal name B is high, the call partner is identified as B.

Next, language specifying processing by the language specifying means will be described in detail.
FIG. 8 shows an example of the speech pattern table 16 for each language. Similar to FIG. 5, the language voice pattern table 16 registers the type of language, the voice pattern, and the appearance probability of the voice pattern corresponding to the language type.
In the above table, a plurality of types of speech patterns that are considered to be answered first by the other party are registered in the language-specific voice pattern table in the same manner as the other party's individual voice pattern table. deep.
Then, as in the case of specifying the individual, the speech pattern of the speech that the call partner first uttered and the speech patterns of various languages registered in advance are collated in the order of appearance probability, and the language type with the highest match rate Determine.
FIG. 9 is a flowchart showing the language specifying process.
First, the first voice of the calling party is captured (step S1), and the first voice pattern A is compared with the voice patterns in the language-specific voice pattern table. That is, as in the case of the individual specifying process, the speech pattern B is extracted from the speech pattern table by language in descending order of appearance rate, the speech patterns A and B are compared, and the hit rate is stored in the language speech pattern hit rate registration means ( Steps S2, S3).
When the comparison is completed, the pattern hit rate registered in the language / speech pattern hit rate registering means is read, and a search is made for a hit rate exceeding a certain value (steps S4 and S5).
If there is a hit rate exceeding a certain value, the one with a high hit rate is identified as the language spoken by the call partner (steps S6 and S7). If no hit rate is found, it is determined that the language cannot be specified, and the process is terminated (step S8).

The appearance probability of the above-mentioned voice pattern table for the other party is updated every time a call is received from the other party, and the voice that is not registered in the table is emitted as the first voice by the other party or a new call is made. When a call is received from the other party, these voices are additionally registered in the table. Hereinafter, the appearance probability update process and the voice pattern addition process will be described with reference to (i)-(xi) of FIG.
(i) Assume that the other party answers, for example, as (a) to (b) as the first voice.
(ii) When the first voice pattern is obtained, the first voice pattern is provisionally registered. For example, in the case of (a), the “high” voice pattern x is provisionally registered. In the case of (b), the voice pattern x of “Hallo” is provisionally registered.
(iii) The temporarily registered voice pattern x is collated with the voice pattern registered in the voice pattern table for each caller shown in FIG.
Here, as shown in FIG. 10, the collation of the voice pattern x and the voice pattern table by call partner in FIG. 5 is performed in the order of the voice pattern having the highest appearance probability to the voice pattern having the lowest appearance probability.
(iv) If there is a voice pattern with a hit rate exceeding a predetermined threshold by the above collation, the call partner who accepted the call as described above is assumed to have the personal name corresponding to the voice pattern, and its appearance probability Update. For example, as a result of the collation, if the hit rate between the voice pattern B1 and the voice pattern x in FIG. 5 exceeds a predetermined threshold, it is determined that the other party receiving the call is “Mr. B” and the voice pattern B1 Is increased by 1, and the appearance probability of the voice pattern B1 of “Mr. B” is calculated and updated. Here, the voice pattern x may be replaced with the existing voice pattern B1. In this way, the voice pattern can be updated to the latest one.
(v) When the other party is specified as described above, the language type information table 17 is referred to as described above, the language type of the other party is specified, and the specified language type is stored in the language type information registration unit 15. sign up.

(vi) On the other hand, if there is no voice pattern whose hit rate exceeds the threshold, as described above, the voice pattern x is compared with the language-specific voice pattern registered in the language-specific voice pattern table, and Identify the spoken language.
(vii) If the language spoken by the call partner can be specified, the specified language type is registered in the language type information registration unit 15.
(viii) If the language spoken by the other party cannot be specified, the call is forwarded to a person who can handle multiple languages.
(ix) If the language spoken by the other party is specified, and the corresponding person knows the individual name of the other party through a call with the other party, the individual name of the other party Check if it is registered in the voice pattern table.
(x) If the other party has already been registered in the voice pattern table for each caller, a new voice pattern x is registered in the corresponding personal name column of the caller-specific voice pattern table, and the voice pattern in that column appears. Update the probability. For example, as shown in FIG. 10, when the other party is C, the voice pattern x is added as C4 to the voice pattern of the other party C, and the appearance probability is calculated and registered. Also, the appearance probability of the voice patterns C1 to C3 is updated.
(xi) If the other party is not registered in the voice pattern table for each caller, as shown in FIG. 10, the voice pattern x is registered in the caller-specific voice pattern table in association with the specified caller name. To do. Also, the appearance probability is registered (in this case, the appearance probability is 100%).

  In addition, the language voice pattern table is obtained by previously registering a voice pattern considered to be the first caller for each language type together with an appearance probability. Although it is not assumed that the appearance probability of the table is updated every time is specified, the language type can be specified as described in FIG. If the identified language type is already registered in the language-specific speech pattern table, the speech pattern is newly registered in the corresponding language type column of the language-specific speech pattern table, and the appearance probability of the speech pattern in that column is updated. You may do it.

It is a figure which shows the system configuration | structure of the Example of this invention. It is a functional block diagram of the processing apparatus of the Example of this invention. It is a flowchart which shows the whole process of the Example of this invention. It is a figure which shows an example of a language classification information table. It is a figure which shows an example of the audio | voice pattern table classified by telephone call partner. It is a flowchart which shows a person specific process. It is a figure which shows the example of registration of a pattern hit rate registration means. It is a figure which shows an example of the audio | voice pattern table according to language. It is a flowchart which shows a language specific process. It is a figure explaining the update of the appearance probability of the voice pattern table classified by call partner, and the additional registration of a voice pattern.

Explanation of symbols

10 Voice pattern registration means 11 Voice pattern storage unit
DESCRIPTION OF SYMBOLS 12 Individual identification means 13 Voice pattern table according to call party 14 Language identification means 15 Language type information registration part 16 Voice pattern table according to language 17 Language type information table 18 Speech recognition means 19 Speech recognition dictionary according to language 20 Translation means 21 Display / speech output Means 22 Language-specific translation dictionary 23 Telephone transfer means 100a exchange 100b exchange 101 public network 102 processing device 103a telephone 103b telephone

Claims (2)

A telephone reception translation system that receives a call from a customer, translates the language spoken by the other party of the received telephone, and talks to the other party.
Voice pattern storage means for storing the first voice uttered by the calling party, the appearance probability of the voice and the number of appearances in association with the personal name;
Corresponding to a pre-registered language type, a language type storage means for storing a voice pattern for each language type that the call partner is supposed to respond to first, and an appearance probability that the voice pattern appears;
When receiving a call, the first voice pattern issued by the call partner and the voice pattern registered in the voice pattern storage means are collated in the order of appearance probability, and the individual specifying means for specifying the call partner;
When an individual is specified by the individual specifying means, a language type extracting means for specifying the language of the other party with reference to language type information storing a language type corresponding to the specified personal name, and the specified language Translation means for translating the language spoken by the other party according to the type,
A language type specifying means for specifying the language of the call partner by comparing the first voice pattern emitted by the call partner and the voice pattern stored in the language type storage means in the order of appearance probability when the call partner cannot be specified;
A telephone reception translation system comprising: language type registration means for registering in the voice pattern storage means the language type information spoken by the specified caller in correspondence with the personal name . A program that receives a call from a customer and causes a computer to execute a telephone reception translation process for translating the language spoken by the other party of the received telephone,
When receiving a call, the first voice uttered by the other party, the appearance probability and the number of appearances of the voice are stored in the storage means with reference to the voice pattern storage means stored in association with the personal name . The process of matching the voice pattern with the first voice pattern emitted by the other party to identify the other party,
When an individual is identified by the above process, a process of extracting a language type for identifying a language of a call partner with reference to language type information storing a language type corresponding to the specified personal name;
A process of translating the language spoken by the other party according to the specified language type ,
A language type memory that stores a speech pattern for each language type that the call partner is supposed to respond to first, and an appearance probability of the appearance of the speech pattern in association with a pre-registered language type when the other party cannot be specified The language type specifying process for specifying the language of the other party by comparing the first voice pattern emitted by the other party with the voice pattern stored in the language type storage unit in the order of appearance probability, with reference to the means,
A telephone reception translation processing program that causes a computer to execute processing for registering language type information spoken by a specified call partner in the voice pattern storage means in association with an individual name .

Images