JP2021078012A - Answering machine determination device, method and program - Google Patents

Abstract

To solve the problem that, when a phone call is made with automatic voices, it cannot be discriminated whether an answer at a call incoming side consists of a human voice or an answering machine message.SOLUTION: An answering machine response determination device comprises: speech data acquisition means which acquires speech data of a phone; speaker overlapped feature amount extraction means which calculates an overlap degree of both speakers in the speech data; clustering feature amount extraction means which calculates a clustering feature amount of the speech data; and determination means which determines whether or not a response is an answering machine message generated by machine learning while using the speaker overlapped feature amount, the clustering feature amount and the presence/absence of a response by an automatic answering machine as teacher data. The determination means determines whether or not the speech data are the response by the answering machine message on the basis of the speaker overlapped feature amount and the clustering feature amount extracted from the speech data.SELECTED DRAWING: Figure 2

Description

The present invention relates to an answering machine determining device, method, and program for determining whether or not the answering machine is answered by an answering machine when an automatic call is made using automatic voice.

If the user does not perform an off-hook operation for an incoming call for a certain period of time, the caller automatically answers, plays a predetermined preset answering machine message, and then prompts the caller to leave a message. Telephones with an answering machine function that can record the message of the above and the function that can set the answering machine service through the communication network are widespread.

However, when the answering machine answers, the caller can leave his message to the other party, but the call charge is charged and the original purpose of making the call by telling the other party the desired message is There was the problem that it could not be achieved.

In order to solve this problem, in Patent Document 1, the telephone device on the calling side measures the answering machine response time of the telephone of the called destination and stores it together with the telephone number of the called destination. A technology is disclosed that prevents unnecessary payment of call charges by automatically disconnecting the call immediately before the memorized answering machine response time is reached.

JP-A-2017-11623

However, in the technique of Patent Document 1, it is possible to prevent unnecessary payment of call charges by automatically disconnecting from the second call, but when making the first call to the other party, it is not possible to make an answering machine. It cannot be prevented. In addition, when the calling side makes a call by automatic voice by automatic calling by a computer program or the like, there is a problem that it is not possible to determine whether the response received by the called side is a human voice or an answering machine message. there were.

If you can't determine whether the callee's phone was an answering machine when it was off-hook, or if a person answered the call but did not respond to the message given by automatic voice, call immediately. I can't decide whether to fix it or call it again after a while. For this reason, there is a problem that the automatic voice telephone unnecessarily repeats the outgoing call operation, such as calling back many times immediately even though the telephone is out of the office.

Therefore, in the present invention, after the call is completed, the call data is acquired from the automatic voice telephone, at least the speaker overlapping feature amount and the clustering feature amount are extracted from the call data, and the determination unit generated by machine learning is used. The purpose is to prevent unnecessary outgoing call operations after that by determining whether or not the answer is from an answering machine, and to be able to take appropriate measures.

In the present invention, a call data acquisition means for acquiring telephone call data, a speaker duplication feature amount extraction means for calculating the degree of duplication of both speakers of the call data, and a clustering feature amount for call data are calculated. Using the clustering feature amount extraction means, the speaker duplication feature amount, the clustering feature amount, and the presence / absence of a response by the answering machine as teacher data, it is determined whether or not the answer is made by the answering machine generated by machine learning. It has a determination means, and the determination means determines whether or not the call data is answered by the answering phone based on the speaker overlapping feature amount extracted from the call data and the clustering feature amount. A response determination device can be provided.

Further, the answering machine answering machine has a call time feature amount extracting means for extracting the feature amount of the call time of the call data, and the determination means further uses the feature amount of the talk time as the teacher data.

The call time feature is an energy statistic of the call voice.

The answering machine answer determination device according to the present invention further includes a voice recognition means for converting call data into text data and a response detecting means for detecting a response by an answering machine from the text data by machine learning and calculating a text feature amount. It further has a text feature amount extracting means, and the determination means further uses the text feature amount as teacher data.

Furthermore, machine learning is performed using machine-synthesized sound data, human voice data, and a mixed Gaussian histogram as teacher data, and the probability that the answering voice is machine-synthesized sound data among the call data is determined by machine-synthesized sound. A machine-synthesized sound feature amount generating means for generating as a feature amount is further provided, and the determination means provides an answering machine answering determination device that further uses the machine-synthesized sound feature amount as teacher data.

The answering machine answer determination method according to the present invention includes a call data acquisition step for acquiring call data by telephone, a speaker duplication feature amount extraction step for calculating the degree of duplication of both speakers of the call data, and clustering of call data. Generated by machine learning using the clustering feature amount extraction step for calculating the feature amount, the talk time feature amount, the speaker overlapping feature amount, the clustering feature amount, and the presence / absence of a response by the answering phone as teacher data. Provided is an answering machine answering determination method having a determination step for determining whether or not the answering is by an answering machine by the determination unit.

Further, the answering machine answering machine to be executed by the computer as the answering machine answering machine of the present invention is a speaker that calculates the degree of duplication of both speakers in the call data acquisition step for acquiring telephone call data and the call data. Teachers of the duplicate feature extraction step, the clustering feature extraction step for calculating the clustering feature of call data, the talk time feature, the speaker duplicate feature, the clustering feature, and the presence or absence of an answering machine response. Provided is an answering machine answering machine answering determination program for executing a determination step of determining whether or not a response is made by an answering machine by a determination unit generated by machine learning using the data.

According to the present invention, it is possible to acquire a predetermined feature amount from the call data and determine whether or not the call destination telephone is a response by an answering machine. Therefore, specifically, it is determined that the call is an answering machine. If this is the case, wait a few hours and try to call again during business hours if it is likely that you are at home or at a store. Therefore, if it is determined that the telephone is immediately disconnected, the telephone can be set to call back immediately.

FIG. 1 is a block diagram showing an example of a hardware configuration diagram of the answering machine answering machine answering machine according to the present invention. FIG. 2 is a functional block diagram of the answering machine answering machine answering machine 1 according to the first embodiment of the present invention. FIG. 3 is a functional block diagram of the answering machine answering machine answering machine 1 according to the second embodiment of the present invention. FIG. 4 is a graph showing the distribution of seconds in the case of human response. FIG. 5 is a graph showing the distribution of seconds in the case of a machine response of an answering machine. FIG. 6 is a flowchart showing a process of generating a classifier by machine learning for determining whether or not the answering machine has answered on the called party side using the teacher data.

Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings. In the present specification and drawings, components having substantially the same function and configuration are designated by the same reference numerals, and duplicate description will be omitted.

FIG. 1 is a block diagram showing an example of a hardware configuration diagram of the answering machine answering machine answering machine according to the present invention. The hardware configuration of the answering machine answering machine 10 which is the computer device shown in FIG. 1 can be realized mainly by the computer device. The answering machine answering machine 10 determines whether or not the answering machine is answered by extracting various feature quantities from the call data received from the automatic voicemail 20 and executing an answering machine answering machine determination program for determining the answering machine answer. Make a judgment.

The answering machine answering machine answering machine 10 has an answering machine answering machine answer determination unit generated by machine learning the call data, various feature quantities, and the presence or absence of an answering machine answer as teacher data, and automatically performs new call data. When it is received from the voice telephone 20, various feature quantities are extracted, and it is determined whether or not the response of the other party is the answering machine.

As shown in FIG. 1, the computer forming the answering machine answering machine 10 includes a CPU 11, a communication interface 12, a ROM 13, a RAM 14, a hard disk drive 15, an input / output interface 16, and a display unit 17 connected to the input / output interface 16. The pointing device 18 and the keyboard 19 are connected to the bus. Further, an external storage device 20 such as a USB memory can be connected to the input / output interface 16.

The display unit 17 is, for example, a display device such as a liquid crystal display. The pointing device 18 is, for example, a mouse or a trackball.

When a series of processes are executed by a program, for example, a call data acquisition unit, a speaker overlapping feature amount extraction unit, a clustering feature amount extraction unit, a talk time feature amount extraction unit, a text feature amount extraction unit, and a machine-synthesized sound feature The amount extraction unit and the determination unit are stored in the ROM 13 or the hard disk drive 15 as an answering machine answer determination program, and are executed by the CPU 11 to execute various functions. It should be noted that the installation by connecting an external storage device 20 such as a USB memory in which the answering machine answering judgment program is stored to the input / output interface 16, the answering machine answering judgment program is installed from the network 12 to the computer, and the device main body. It may be configured in a state preliminarily incorporated in, for example, a ROM 13 in which an answering machine answer determination program is recorded.

FIG. 2 is a functional block diagram of the answering machine answering machine answering machine 1 according to the first embodiment of the present invention. The answering machine answering machine answering machine 1 is composed of an automatic voice telephone 20 and an answering machine answering machine 10. The automatic voice telephone 20 is a telephone that can make a call without a human being by automatically calling a predetermined telephone number and listening to a predetermined message with an artificially generated voice to the other party. For example, by registering the telephone reservation message required for making a reservation call to the store in the automatic voice telephone 20 in advance and appropriately combining it with the telephone reservation message such as the desired date and time of reservation and the number of people, the telephone number of the store to be reserved Is read, the call is automatically made, the generated message for telephone reservation is read aloud by artificial voice, and the other party is inquired whether the reservation is possible. The automatic voice telephone 20 may be configured by installing a computer program that makes a call by referring to a telephone number stored in a database on a computer such as a server, and is configured by a plurality of computer devices. You may.

The answering machine answering machine 10 is generated by receiving the call data recorded by the automatic voice telephone 20 via a communication network such as the Internet, extracting various feature quantities from the call data, and performing machine learning using the teacher data. It is a device that determines whether or not the other party has responded with an answering machine message based on the determination unit.

The automatic voice telephone 20 and the answering machine answering machine 10 are shown as separate devices here, but the present invention is not limited to this, and the automatic voice telephone 20 and the answering machine answering machine 10 are configured as one device. You may. Further, each function constituting the automatic voice telephone 20 and the answering machine answering machine 10 may be configured as an independent device.

The answering machine answering machine 10 includes a call data acquisition unit 101, a speaker overlapping feature amount extraction unit 111, a clustering feature amount extraction unit 113, and a determination unit 121.

The call data acquisition unit 101 acquires the call data by receiving the call data from the automatic voice telephone 20 that the automatic voice telephone 20 automatically makes a call and makes a call without human hands. The data format of the call data may be recorded in a dual channel of 2 channels consisting of the voice of the caller and the voice of the callee. For example, the wav format is used, but the data format is particularly limited to a specific data format. Absent.

The speaker duplication feature amount extraction unit 111 calculates the degree of duplication of both speakers in the call data. Specifically, since the call data is recorded in dual channels on the caller side and the callee side, the speaker duplication feature amount extraction unit 111 detects the section talking on both channels and duplicates it. Judge the degree. The speaker duplication feature amount extraction unit 111 has, for example, a voice section detector (VAD: Voice Activity detection), extracts talk sections on both the caller side and the callee side, and calculates the degree of overlap. To do. For example, a linear model of MFCC (Mel-frequency cepstrum coefficients) may be used as the voice interval detector. The speaker overlap feature amount extraction unit 111 performs the following calculation to calculate the overlap time (overlap) of the talking section between the caller side and the callee side. Shello is the section spoken by the caller, Srestaurant is the section spoken by the called party, and | S | is the length of seconds in that section.

The clustering feature amount extraction unit 113 calculates the clustering feature amount from the voice data of the channel only on the called party side in the call data. The clustering feature amount extraction unit 113 performs clustering so as to be classified into a response by a machine such as a response by an answering machine, a response by a human, and a characteristic voice pattern of each response. Specifically, a feature called BoAW (Bag of Audio Words) is calculated. The clustering feature amount extraction unit 113 first calculates the MFCC feature amount for the received call data.

Specifically, by setting one frame to 0.025 seconds and shifting by 0.01 seconds, 100 frames are generated per second, the 40-dimensional MFCC features in each frame are regarded as high-dimensional points, and the set of points is k. Generate clustering by applying to clustering of -means. k-means generates clustering by minimizing the following equation. x is the point in each frame, Si is the set of points included in the i-th cluster, μi is the center of the cluster, and S is the set of all clusters.

For example, if 100 10-second call data are prepared, it is assumed that 10 × 100 × 100 = 100,000 points exist, and k-means clustering is applied to this to generate 100 clusters. Next, the MFCC feature amount of the call data for which the answering machine answer determination is performed is calculated, and the distance to which clustering the MFCC feature amount of each frame is close is calculated. By allocating the frame to the nearest clustering, a clustering histogram is generated and the clustering features are extracted. That is, the clustering feature amount extraction unit 113 extracts the clustering feature amount by generating clustering from the MFCC feature amount which is the voice feature amount for the voice data on the called party side and further generating a clustering histogram.

The determination unit 121 is generated by machine learning using various feature amounts extracted from the call data and the presence / absence of a response by the answering machine as teacher data. By inputting the speaker overlapping feature amount and the clustering feature amount extracted from the call data for which it is desired to judge whether or not the answer was made by the answering machine into the judgment unit 121 generated by machine learning, the judgment unit 121 can perform the call data. In, it is determined whether or not there is a response by the answering machine.

The determination unit 121 is composed of a classifier that performs binary classification when a feature amount extracted from the target call data is input. Here, it is classified into two values, a machine response and a human response, which are responses by an answering machine. As the classifier, for example, logistic regression, random forest, SVM (support vector machine), etc. may be used, and any of them may be used, but when the teacher data is small, for example, when it is less than 10000 voice data, logistic regression, In the case of more than that, SVM or the like may be used properly.

For example, logistic regression shows the best performance when there is little teacher data, and it is conditional that it is classified into machine response C1 which is the response by the answering machine from the input feature x under the parameter of the weight w by the following formula. Calculate the probability.
P (C1 | x; w) = σ (w ^T X + w ₀ )

When using a random forest as a classifier, the calculation is performed based on _{the result fk of each decision tree.}

When the amount of data reaches a certain level or more, for example, 10,000 voice data or more, SVM can classify by replacing it with logistic regression when using higher performance as a classifier. By using the kernel trick, the nonlinear classification ability can be realized by drawing a hyperplane on a higher dimension than originally intended. The boundary surface is determined by the following formula. (Zi, yi) are the features and labels of the existing i-th teacher data, and w and b are the weights to be learned.

FIG. 3 is a functional block diagram of the answering machine answering machine answering machine 1 according to the second embodiment of the present invention. In the second embodiment, features other than the speaker overlapping feature amount and the clustering feature amount are also extracted and used for determining whether or not the response was made by an answering machine. A detailed description of the same configuration as that of the first embodiment will be omitted.

The answering machine answering machine answering machine 1 is composed of an automatic voice telephone 20 and an answering machine answering machine 10. The automatic voice telephone 20 is a telephone that can make a call without a human being by automatically calling a predetermined telephone number and listening to an artificially generated voice of a predetermined message to the other party. The answering machine answering machine 10 is generated by receiving the call data recorded by the automatic voice telephone 20 via a communication network such as the Internet, extracting various feature quantities from the call data, and performing machine learning using the teacher data. It is a device that determines whether or not the other party has responded with an answering machine message based on the determination unit.

In the present embodiment, the answering machine answer determination device 10 includes a call data acquisition unit 101, a speaker overlapping feature amount extraction unit 111, a clustering feature amount extraction unit 113, a talk time feature amount extraction unit 115, and a text feature amount extraction unit. It has 117, a machine-synthesized sound feature amount extraction unit 119, and a determination unit 121.

The call data acquisition unit 101 acquires the call data by receiving the call data from the automatic voice telephone 20 that the automatic voice telephone 20 automatically makes a call and makes a call without human hands. In addition, the speaker duplication feature amount extraction unit 111 calculates the degree of duplication of both speakers in the call data. The clustering feature amount extraction unit 113 calculates the clustering feature amount from the voice data of the channel only on the called party side in the call data. Since the speaker overlapping feature amount extraction unit 111 and the clustering feature amount extraction unit 113 have the same configuration as that of the first embodiment, detailed description thereof will be omitted here.

The call time feature amount extraction unit 115 extracts the feature amount of the call time of the call data. For example, the call time itself in the call data may be used as a feature quantity. Further, in another example, the square of the talk time may be used as the feature amount of the talk time. As for the call time when answering the answering machine, the recording of a fixed template is often used, so the call ends at the same time. Therefore, by using the square of the talk time as a feature quantity, a quadratic function can be expressed to capture the peak when the answering machine is answered.

そして、ナイキスト周波数までのパワースペクトルを足し合わせることで、エネルギー統計量を算出する。

Further, the talk time feature amount extraction unit 115 may extract the energy statistics of each frame of voice as the talk time feature amount. In this case, the noise over the entire talk time is measured by calculating the energy statistics of each frame of voice. If the energy statistic is high, it means that there is a lot of noise in the background, and if it is low, it means that there is not much noise in the background. Specifically, the following formula is calculated as an energy statistic. The audio signal is x (t), divided into 0.025 line frames, and the window function w (t) is applied to convert the signal X [t, f] by short-time Fourier transform. t is time and f is frequency.

Then, the energy statistic is calculated by adding the power spectra up to the Nyquist frequency.

The talk time feature amount extraction unit 115 may extract only the energy statistic and the talk time feature amount, and as the optimum embodiment, the talk time feature amount extraction unit 115 extracts two feature amounts of the energy statistic and the square of the talk time. You may. Two feature quantities, an energy statistic and a talk time, may be extracted.

The text feature extraction unit 117 includes a voice recognition unit 1171 that converts text data, and a response detection unit 1172 that detects a response from an answering machine by machine learning from the text data and calculates a text feature amount. The voice recognition unit 1171 performs voice recognition of the voice data on the called party side among the acquired call data and converts it into text.

The response detection unit 1172 detects whether or not there is a typical text message of a machine response by an answering machine from the text data that is voice-recognized and generated by the voice recognition unit 1171. For example, it detects frequently used messages such as "I'm away", "Please give me a message", and "It's out of business hours". Specifically, the response detection unit 1172 applies a Japanese pre-learning language model such as BERT (Bidirectional Encoder Representations from Transformer) or XLNet to perform response detection.

The machine-synthesized sound feature amount extraction unit 119 extracts a feature amount for determining whether or not the voice of the called party is a machine-synthesized sound. For example, in the machine-synthesized sound feature amount extraction unit 119, a human A large amount of each of voice and artificially synthesized voice is collected, and they are put into LSTM (Long Short-Term Memory) to be trained so that human voice and synthetic voice can be classified. In addition, a histogram of the mixed Gaussian distribution of speech data is calculated, a histogram in the case of human speech and a histogram in the case of synthetic speech are calculated, and these are machine-learned as teacher data by a neural network to generate a machine-synthesized sound classifier. Generate it.

When the machine-synthesized sound feature amount extraction unit 119 acquires the call data, the machine-synthesized sound feature amount extraction unit 119 extracts the voice data of the channel on the called party side, calculates a histogram of the mixed Gaussian distribution of the voice data, and machine-learns the machine-synthesized sound classifier. Is asked to calculate the probability of whether or not it is a synthetic voice, and the result is extracted as a machine-synthesized voice feature.

The determination unit 121 is generated by machine learning using various feature amounts extracted from the call data and the presence / absence of a response by the answering machine as teacher data. By inputting the feature amount extracted from the call data for which it is desired to determine whether or not the answer was received by the answering machine into the determination unit generated by machine learning, the determination unit 121 indicates whether or not there was a response by the answering machine in the call data. Is determined. In the present embodiment, the speaker overlapping feature amount, the clustering feature amount, the talk time feature amount, the text feature amount, and the machine-synthesized sound feature amount are input and the determination is performed.

In the second embodiment, five feature amounts of speaker overlapping feature amount, clustering feature amount, talk time feature amount, text feature amount, and machine-synthesized sound feature amount are input, but the text feature amount or the machine The determination may be made by inputting one of the synthetic sound features and the other three features. Further, the determination may be made by inputting three feature amounts, that is, the speaker overlapping feature amount, the clustering feature amount, and the talk time feature amount. Processing can be speeded up by using only these three feature quantities.

FIG. 4 is a graph showing the distribution of seconds in the case of human response. As shown in the figure, it can be seen that when a person comes out on the called side, the talk time often ends in a short number of seconds. Probably because when a call is received by automatic answering, the called party tends to hang up immediately when the other party knows that it is an automatic answering (machine).

FIG. 5 is a graph showing the distribution of seconds in the case of a machine response of an answering machine. It can be seen that the peak of the talk time is at 50 to 60 seconds. In the case of a machine response by answering machine, it takes a certain number of seconds because a standard message is played. When making a call with automatic answering, if the called party is a machine answering with an answering machine, there is time to record the message after the answering message, and after the predetermined recording time, the called party disconnects, so the peak time is 50. It is expected to be ~ 60 seconds.

FIG. 6 is a flowchart showing a process of generating a classifier by machine learning for determining whether or not the answering machine has answered on the called party side using the teacher data. First, the call data acquisition unit 101 acquires the answer result data indicating whether or not the answering machine is answered by the called party, and the call data (step S601). Here, since the purpose is to make the classifier perform machine learning, the answer result data is acquired together with the call data as the teacher data.

Next, each feature amount extraction unit extracts each feature amount from the call data (step S602). In the first embodiment, the speaker overlapping feature amount extracting unit 111 and the clustering feature amount extracting unit 113 extract the speaker overlapping feature amount and the clustering feature amount from the call data. In the second embodiment, the speaker overlapping feature amount extracting unit 111, the clustering feature amount extracting unit 113, the talk time feature amount extracting unit 115, the text feature amount extracting unit 117, and the machine-synthesized sound feature amount extracting unit 119 Extract speaker overlapping features, clustering features, talk time features, text features, and machine-synthesized sound features.

Next, a classifier used in the determination unit is generated by machine learning (step S603). By machine learning the various features extracted from the call data and the presence or absence of a response by the answering machine as teacher data, a classifier for determining whether or not the response was by the answering machine is generated. Examples of the machine learning method include logistic regression, random forest method, and support vector machine, and any of them may be used. As the feature amount, in the first embodiment, the speaker overlapping feature amount and the clustering feature amount, and in the second embodiment, the speaker overlapping feature amount, the clustering feature amount, the talk time feature amount, and the text feature amount. Amount, machine-synthesized sound feature amount. The classifier generated from the teacher data determines whether or not the call data to be determined is a response by an answering machine.

1 Answering machine answering machine 10 Answering machine answering machine 20 Automatic voice telephone

Claims (7)

Call data acquisition means to acquire call data by telephone,
A speaker duplication feature amount extracting means for calculating the multiplicity of both speakers of the call data, and
A clustering feature extraction means for calculating the clustering feature of the call data,
Using the speaker overlapping feature amount, the clustering feature amount, and the presence / absence of a response by an answering machine as teacher data, a determination means for determining whether or not the answer is a response by an answering machine generated by machine learning.
Have,
The determination means is an answering machine answering machine that determines whether or not the call data is answered by an answering machine based on the speaker overlapping feature amount extracted from the call data and the clustering feature amount.
The answering machine answering machine according to claim 1, further comprising a call time feature amount extracting means for extracting a call time feature amount of the call data.
The determination means is an answering machine answer determination device that further uses the feature amount of the talk time as teacher data.
The answering machine answering machine according to claim 2, wherein the talk time feature amount is an energy statistic of the call voice.
The answering machine answering machine according to claims 1 to 3.
Further, it further has a text feature amount extracting means having a voice recognition means for converting the call data into text data and a response detection means for detecting a response by an answering machine from the text data by machine learning and calculating a text feature amount. And
The determination means is an answering machine answer determination device that further uses the text feature amount as teacher data.
The answering machine answering machine according to claims 1 to 4.
Furthermore, machine learning is performed using machine-synthesized sound data, human voice data, and a mixed Gaussian histogram as teacher data, and the probability that the answering voice is machine-synthesized sound data among the call data is determined by machine-synthesized sound. It also has a means for generating machine-synthesized sound features that are generated as features.
The determination means is an answering machine answer determination device that further uses the machine-synthesized sound feature amount as teacher data.
It is an answering machine answer judgment method,
Call data acquisition step to acquire call data by telephone and
A speaker duplication feature amount extraction step for calculating the multiplicity of both speakers of the call data, and
A clustering feature extraction step for calculating the clustering feature of the call data, and
Using the talk time feature amount, the speaker overlapping feature amount, the clustering feature amount, and the presence / absence of a response by an answering machine as teacher data, a response by an answering machine is performed by a determination unit generated by machine learning. Judgment step to determine whether or not
Answering machine answer judgment method having.
An answering machine answering machine that is executed by a computer as an answering machine answering machine.
Call data acquisition step to acquire call data by telephone and
A speaker duplication feature amount extraction step for calculating the multiplicity of both speakers of the call data, and
A clustering feature extraction step for calculating the clustering feature of the call data, and
Using the talk time feature amount, the speaker overlapping feature amount, the clustering feature amount, and the presence / absence of a response by an answering machine as teacher data, a response by an answering machine is performed by a determination unit generated by machine learning. Judgment step to determine whether or not
Answering machine answering machine to execute.

Images