JP2002252705A - Method and device for detecting talker id - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method and device for detecting a talker ID that limits number of recognition objects to recognize a talker's voice and to identify the talker in the case of recognizing and identifying the talker making a phone call. SOLUTION: Upon the receipt of a phone call, a caller number is detected (S1), the detected caller number and telephone numbers in a data storage table are compared (S2) by using the data storage table where each telephone number, each talker ID and each of voice data to identify the talker are recorded as each set, the data storage table including a telephone number in matching with the detected caller number is extracted as an object. If there exist a plurality of extracted objects, the voice data to identify the talker are collated with a phone voice of the caller (S4), and the talker ID with the highest similarity is selected and the talker is identified.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for detecting a speaker ID, and more particularly, to a device for specifying a speaker ID of an incoming call in an answering machine, an automatic telephone answering system, a call center, and the like.

[0002]

2. Description of the Related Art As a method of specifying a caller ID of a caller, for example, a method of using a detecting device for detecting a telephone number of a caller disclosed in Japanese Patent Application Laid-Open No. 10-285286 is widely known. I have. However, it may not be possible to identify the speaker only by detecting the telephone number. For example, in the case of a company or home telephone, since a plurality of people call using the same telephone number, it is not possible to identify any of them. Conversely, when a company has a plurality of telephone lines, each time a call is made, a different line is used, and it cannot be determined that they are the same speaker.

On the other hand, Japanese Patent Laid-Open Publication No. Hei 10-322450 recognizes the voice of a caller, the voice of a speaker who receives a call (such as a receptionist in a company), or both. Thus, a method of specifying a speaker ID is also known. However, when the customer database from which a speaker ID is to be obtained is large, such as in a company reception, the number of recognition targets is very large, and there is a problem that the possibility of erroneous recognition is large. For this reason, since an accurate speaker ID cannot be obtained, customer information such as past order contents, repair request contents, and question contents cannot be accurately obtained, and a problem that a smooth conversation between a sender and a receiver cannot be performed. there were.

[0004]

SUMMARY OF THE INVENTION The present invention has been made in view of the above points, and has as its object to accurately acquire a speaker ID by recognizing speech by narrowing down a recognition target. It was done.

[0005]

According to the first aspect of the present invention, when a telephone call is received, a caller ID is detected, and the telephone number and the speaker ID are detected.
The detected caller number is compared with the telephone number in the data hold table using a data holding table as a set of D and voice data for identifying the speaker, and matches with the detected caller number. A data holding table having a telephone number to be used as a candidate, collating the voice data for identifying the speaker with the telephone voice of the caller, and obtaining a speaker ID having the highest similarity. is there.

According to a second aspect of the present invention, when a call is received, a caller number is detected, and the detection is performed using a data holding table in which a telephone number, a speaker ID, and voice data for identifying the speaker are combined. Comparing the detected caller number with the telephone number of the data holding table, and identifying the speaker as a data holding table having a telephone number that partially matches the detected caller number from the beginning as a candidate. The voice data is compared with the caller's telephone voice to obtain a speaker ID having the highest similarity.

According to a third aspect of the present invention, a caller number is detected when a telephone call is received, and a telephone number, a speaker ID, voice data for identifying the speaker, and ID data for identifying the speaker are combined. Compare the detected caller number and the telephone number of the data holding table using a data holding table as a, and as a candidate a data holding table having a telephone number that matches the detected caller number, ID data for identifying a speaker is compared with a response voice of the receiver to the speaker, and a speaker ID having the highest similarity is obtained.

According to a fourth aspect of the present invention, a caller number is detected when a telephone call is received, and a telephone number, a speaker ID, voice data for identifying the speaker, and ID data for identifying the speaker are combined. The detected caller number is compared with the telephone number in the data hold table using the data hold table as a candidate for the data hold table having a telephone number that partially matches the detected caller number from the beginning. As
ID data for identifying the speaker is compared with a response voice of the receiver to the speaker, and a speaker ID having the highest similarity is obtained.

According to a fifth aspect of the present invention, there is provided a caller number detecting means for detecting a caller number when receiving a telephone call, and a data holding unit comprising a telephone number, a speaker ID, and voice data for identifying the speaker. A data holding means for holding a table, comparing the caller number detected by using the caller number detecting means with a telephone number in the data holding table, and matching the detected caller number. The data holding table having a telephone number is set as a candidate, and the voice data for identifying the speaker is compared with the telephone voice of the caller to obtain a speaker ID having the highest similarity. .

According to a sixth aspect of the present invention, there is provided a caller number detecting means for detecting a caller number when a telephone call is received, and a data holding unit comprising a telephone number, a speaker ID, and voice data for identifying the speaker. Data holding means for holding a table, comparing the caller number detected by using the caller number detecting means and the telephone number of the data holding table, from the top to the detected caller number Assuming that a data holding table having a partially matching telephone number is a candidate, voice data for identifying the speaker is compared with telephone call data of the caller to obtain a speaker ID having the highest similarity. It is what it was.

According to a seventh aspect of the present invention, a caller number detecting means for detecting a caller number when a call is received, a telephone number, a speaker ID, and voice data for identifying the speaker and the speaker are identified. And a data holding means for holding a data holding table as a set of ID data for comparing the caller number detected by using the caller number detecting means and the telephone number of the data holding table, Using the data holding table having a telephone number that matches the detected caller number as a candidate, the ID data for identifying the speaker is compared with the response voice of the receiver to the speaker, and the similarity is determined to be the highest. It is characterized by obtaining a high speaker ID.

[0012] The invention of claim 8 is a caller number detecting means for detecting a caller number when receiving a telephone call, a telephone number, a speaker ID, and voice data for identifying the speaker and the speaker. And a data holding means for holding a data holding table as a set of ID data for comparing the caller number detected by using the caller number detecting means and the telephone number of the data holding table, A candidate is a data holding table having a telephone number that partially matches the detected caller number from the beginning, and compares the ID data for identifying the speaker with the response voice of the receiver to the speaker. The feature is that a speaker ID having a high degree of similarity is obtained.

According to a ninth aspect of the present invention, there is provided a computer-readable recording medium storing a program for causing a computer to execute the speaker ID detecting method according to the first aspect.

According to a tenth aspect of the present invention, there is provided a computer-readable recording medium storing a program for causing a computer to execute the speaker ID detecting method according to the second aspect.

According to an eleventh aspect of the present invention, there is provided a computer-readable recording medium storing a program for causing a computer to execute the speaker ID detecting method according to the third aspect.

According to a twelfth aspect of the present invention, there is provided a computer-readable recording medium storing a program for causing a computer to execute the speaker ID detecting method according to the fourth aspect.

[0017]

(Embodiment 1) FIG. 1 is a block diagram showing a configuration example of a speaker ID detecting apparatus to which the present invention is applied. , Data holding means 3 and collation means. A caller number is detected by a caller number detecting means 1 for a call made from outside such as a public network. The means for detecting the caller ID has already been commercialized and widely used, and a description thereof will be omitted.

FIG. 2 is a diagram showing an example of a data holding table according to the present invention. In FIG.
The data holding table 2a stores a speaker ID such as a customer number, a telephone number, and a voice number for identifying a speaker. Holding as a pair.

As information for identifying a speaker, data (parameters) used in a widely known technique for identifying a speaker called speaker recognition or speaker verification may be prepared. Although details are omitted, for example, there is a long-time spectrum that does not depend on words, or a parameter such as “Hiraki Sesame” that simultaneously measures a specific word and speaker characteristics for each speaker.

FIG. 3 is a flowchart illustrating an example of the speaker ID detection processing to which the present invention is applied. First, the caller number of the caller is obtained (step S).
1) Using the data holding table 2a shown in FIG. 2, a data set that matches the detected telephone number is detected and set as a candidate (step S2). It is determined whether there is one candidate (step S3), and if there is only one candidate (YES)
In this case, the number of speaker IDs is determined to be one, and the process ends. On the other hand, when there are a plurality of data sets having the same telephone number (in the case of NO), these are all set as candidates and collation with the telephone voice is performed (step S4).
The candidate with the highest similarity is detected, and this speaker ID is output (step S5).

As another embodiment of the flow of the speaker ID detection processing shown in FIG. 3, the detected telephone number is compared with the telephone number in the data holding table 2a, and the data of the partially coincident data from the head of the telephone number is compared. Find pairs. For example, when the number of digits of the partial match is 6 digits, the caller number is 012345.
If it is 6789, a data set having a telephone number of 012345xxxx (x is an arbitrary number) in the data holding table 2a is detected, all of them are set as candidates, collation with telephone voice is performed, and And outputs the speaker ID.

As another embodiment of the flow of the speaker ID detection processing shown in FIG. 3, as the ID information for identifying the sender, the name of the sender and the like are stored in the data holding table 2a.
Register with. At the time of verification, the name is recognized by the voice of the recipient. Since the receiver can identify the speaker in advance, such as a company telephone operator, an unspecified speaker-type speech recognition method that cannot identify the speaker, such as a specific speaker-type speech recognition method or a speaker-adaptive speech recognition method. , A voice recognition method having a higher recognition rate can be used. Also, given the natural nature of the conversation between the sender and the recipient, it is unnatural for the recipient to utter the name of the sender just for recognition, and the repetition is usually "Yes, XX." It is desirable to recognize the name of the recipient (XX part) from the utterance like this. Therefore, a continuous speech recognition using a grammar according to a presumed repetition pattern or a recognition method of extracting the XX portion by a method such as word spotting is preferable.

As a procedure of this embodiment, first, a data set that matches the telephone number of the caller is detected. When there are a plurality of data sets, those sets are set as candidates. Then, the utterance of the receiver, "Yes, XX-san," is collated with the candidate name, and the speaker ID of the candidate with the highest similarity is output. In addition, a set of data that partially matches the telephone number of the caller from the beginning is set as a candidate, and then the repetition of the receiver and the name of the candidate are collated to output the speaker ID with the highest similarity. .

(Embodiment 2) FIG. 4 is a block diagram showing a configuration example of a speaker ID detection apparatus to which the present invention is applied.
Reference numeral 4 denotes a receiver voice input unit. The present embodiment relates to a device for allowing a telephone receiving organization of a company such as a call center to answer a call received from a public network or the like.

For a call received from a public network or the like,
First, the caller number detecting means 1 detects a caller number. The caller number detecting means 1 is widely known and will not be described here. The detected caller number is compared with the telephone number of each data set in the data holding table 2a, and a matched telephone number or a data set having a partially matched telephone number from the beginning of the telephone number is detected as a candidate. . If there are a plurality of candidates, first, the voice uttered by the telephone operator is input by the receiver voice input means 4 such as a receiver and sent to the matching means 3.

Since the target voice is a part where the caller's name is repeated, such as "Yes, Mr. XX", it is sufficient to recognize only the head part of the conversation. Further, the receiver may press the input switch at the time of reciting and recognize only when the switch is pressed. When a switch is used, a temporal target section to be subjected to voice recognition in a long conversation voice can be accurately limited, so that more precise collation can be performed. As a result of the recognition, a speaker ID having the highest similarity is output.

(Embodiment 3) FIG. 5 is a block diagram showing a configuration example of a speaker ID detecting apparatus to which the present invention is applied.
Reference numeral 5 denotes data presentation means. This embodiment relates to a telephone for home use or the like. The data holding means 2 includes:
Information for presenting a speaker ID such as a speaker ID, a telephone number, voice data for identifying the speaker, and the name of the caller is stored as a set. Here, the voice data for identifying the speaker is speaker-specific data used in a well-known speaker verification technique. For example, it is data such as a long-time spectrum.

First, the caller number detecting means 1 detects the caller number, and detects data corresponding to this number from the data holding means 2. When there is one matching data, the speaker ID is output to the data presenting means 5. If there is a plurality of matching data, the voices of the callers are collated by the collation means 3, the data of the speaker having the highest similarity is selected, and the speaker ID is output. I do.

The data presenting means 5 performs visual presentation such as a display. In this case, the data holding means 2
The presentation information of the speaker ID held in is the text information such as the name and nickname of the speaker. Data presentation means 5
May be output as a voice such as "Call from XX". In this case, the speaker ID presentation information of the data holding unit 2 is output as text information by a text-to-speech conversion speech synthesizer. Alternatively, an audio waveform may be held as it is as data presentation information, and the audio waveform may be reproduced by a reproducing device. FIG. 6 is a diagram showing a data holding table 2a to which speaker ID presentation information is added.

Another example of the embodiment shown in FIG. 5 is shown. The basic operation is the same as in the third embodiment. The difference is that a telephone number that partially matches the telephone number detected by the caller number detecting means 1 from the beginning is detected. The number of digits of the partial match varies depending on the purpose, but when used within the same company, if the main purpose is to identify, a partial match of approximately eight digits from the beginning should be detected. Good. In addition, if the main purpose is to identify a call from a pay phone on the go, it is sufficient to detect a partially coincident telephone number of about six digits from the beginning corresponding to the area code. . In the following processing, similar to the embodiment shown in FIG. 5, when there is one partially matched data, the speaker ID is output to the data presenting means 5 as it is, and when there are a plurality of data, speaker verification is performed. Speaker I with the highest similarity
D is output.

[0031]

According to the first, fifth and ninth aspects of the present invention, the speaker ID of the caller can be accurately specified. That is,
Just by matching the caller numbers, it is not possible to distinguish between speakers using the same phone number, such as family members and businesses, but it is possible to identify them and to correctly identify the caller. In addition, since the range to be compared is only the part where the caller numbers match, it is possible to perform the comparison with a high recognition rate. Further, since the caller can be specified accurately, it becomes possible to precisely access information unique to the caller (such as past order contents).

According to the second, sixth and tenth aspects of the present invention, the speaker ID of the caller can be specified accurately. When there are multiple telephone lines, such as in a company or office, and the same speaker is called with a different caller ID, the caller could not be identified. By collating a range that partially matches from the beginning of the number, the speaker ID can be accurately specified.

According to the third, fourth, seventh, eighth, eleventh and twelfth aspects of the present invention, the speaker of the voice to be collated can be specified in advance like a telephone operator. In this case, a voice recognition method having a high recognition rate can be used, and a speaker ID can be obtained by precise collation.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration example of a speaker ID detection device to which the present invention is applied.

FIG. 2 is a diagram illustrating an example of a data holding table according to the present invention.

FIG. 3 is a flowchart illustrating an example of a speaker ID detection process to which the present invention is applied.

FIG. 4 is a block diagram illustrating a configuration example of a speaker ID detection device to which the present invention is applied;

FIG. 5 is a block diagram illustrating a configuration example of a speaker ID detection device to which the present invention is applied;

FIG. 6 is a diagram showing a data holding table to which speaker ID presentation information is added.

[Explanation of symbols]

1 ... Caller ID detecting means, 2 ... Data holding means, 2a ...
Data holding table, 3 ... collation means, 4 ... receiver voice input means, 5 ... data presentation means.

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) H04M 11/00 302 G10L 3/00 551A

Claims (12)

[Claims] 1. A caller number is detected when a telephone call is received, and the detected caller number is detected using a data holding table in which a telephone number, a speaker ID, and voice data for identifying a speaker are combined. And the telephone number of the data holding table, and using the data holding table having a telephone number that matches the detected caller number as a candidate, voice data for identifying the speaker and telephone sound of the caller And a speaker ID having the highest similarity is obtained. 2. A caller number is detected when a telephone call is received, and the detected caller number is detected using a data holding table in which a telephone number, a speaker ID, and voice data for identifying the speaker are combined. And the telephone number in the data holding table, and using the data holding table having a telephone number partially matching the detected caller number from the beginning as a candidate, voice data for identifying the speaker and the caller A speaker ID having the highest similarity is obtained by collating with the telephone voice of the present invention. 3. A data holding table for detecting a caller number when receiving a telephone call, and assembling a telephone number, a speaker ID, voice data for identifying the speaker, and ID data for identifying the speaker. And comparing the detected caller number with the telephone number in the data holding table, and identifying the speaker as a candidate data holding table having a telephone number that matches the detected caller number. I for
A method for detecting a speaker ID, comprising: comparing D data with a response voice of the receiver to the speaker to obtain a speaker ID having the highest similarity. 4. A data holding table which detects a caller number when receiving a telephone call, and includes a telephone number, a speaker ID, voice data for identifying the speaker, and ID data for identifying the speaker. And comparing the detected caller number with the telephone number in the data holding table, and selecting the data holding table having a telephone number partially matching the detected caller number from the beginning as a candidate, A speaker ID having the highest similarity is obtained by comparing ID data for identifying the ID with a response voice of the receiver to the speaker. 5. A caller number detecting means for detecting a caller number when a telephone call is received, and data holding a data holding table comprising a set of a telephone number, a speaker ID, and voice data for identifying the speaker. Holding means for comparing a caller number detected by using the caller number detecting means with a telephone number in the data holding table, and having a telephone number matching the detected caller number. A speaker ID detection apparatus, wherein a speaker ID having the highest similarity is obtained by collating voice data for identifying said speaker with a telephone voice of a caller using a holding table as a candidate. 6. A caller number detecting means for detecting a caller number when a telephone call is received, and data holding a data holding table comprising a set of a telephone number, a speaker ID, and voice data for identifying the speaker. Holding means for comparing a caller number detected by the caller number detecting means with a telephone number in the data holding table, and a telephone number partially coincident with the detected caller number from the beginning. A speaker ID having the highest similarity by comparing the voice data for identifying the speaker with the data of the telephone voice of the caller as a candidate for a data holding table having Detection device. 7. A caller number detecting means for detecting a caller number when a call is received, a telephone number, a speaker ID, voice data for identifying the speaker, and an ID for identifying the speaker.
Data holding means for holding a data holding table as a set of data, comparing the caller number detected by using the caller number detecting means with the telephone number of the data holding table, and A data holding table having a telephone number that matches the caller number
A speaker ID detection device, wherein ID data for identifying the speaker is compared with a response voice of a receiver to the speaker to obtain a speaker ID having the highest similarity. 8. A caller number detecting means for detecting a caller number when receiving a telephone call, a telephone number, a speaker ID, voice data for identifying the speaker, and an ID for identifying the speaker.
Data holding means for holding a data holding table as a set of data, comparing the caller number detected by using the caller number detecting means with the telephone number of the data holding table, and The ID data for identifying the speaker and the response voice of the receiver to the speaker are compared with each other as a candidate for a data holding table having a telephone number that partially matches the caller number from the beginning. A speaker ID detection device characterized by obtaining a high speaker ID. 9. A computer-readable recording medium on which a program for causing a computer to execute the speaker ID detection method according to claim 1 is recorded. 10. A computer-readable recording medium on which a program for causing a computer to execute the speaker ID detection method according to claim 2 is recorded. 11. A computer-readable recording medium on which a program for causing a computer to execute the speaker ID detection method according to claim 3 is recorded. 12. A computer-readable recording medium storing a program for causing a computer to execute the speaker ID detection method according to claim 4.