Techniques for identifying unwanted calls

Abstract

Method for identifying unwanted calls to a call-capable terminal (12) of a subscriber in a communication network, comprising receiving audio signals of the call that are recorded by an external recording device and sent to an identification server (18) of the communication network (10), wherein an algorithm for identifying unwanted calls is implemented on the identification server, which algorithm is programmed to carry out a voice analysis wherein a dialect corresponding to the location of the subscriber is stored as a voice profile, wherein the voice analysis is set up to extract speech as audio signals from the call information data and to compare these with the known voice profiles and to identify an unwanted call by means of the comparison, wherein it indicates an unwanted call if the dialect of the caller deviates too much from the local or national dialect and After an unwanted call has been identified by the identification server, a technical countermeasure is initiated.

Description

[0001] The present invention relates to the technical field of identifying incoming calls to a subscriber. Specifically, it concerns techniques that can analyze and classify calls and assign them to a specific group, for example, into a group of desired or undesired calls. Specifically, the invention relates to a method for identifying undesired calls, as well as a corresponding communications network and an identification server.

[0002] Fraudulent phone calls are becoming increasingly popular. Such fraudulent phone calls may also benefit from innovative automation processes that can make such calls partially autonomously or assist humans in making them. Furthermore, the increasing prevalence of so-called flat-rate plans makes such phone calls increasingly lucrative for fraudsters. Not only the elderly, but essentially anyone can become a victim of fraud and potentially lose large sums of money.

[0003] Currently, there is no effective defense against fraudulent phone calls, except for the subscriber, i.e., the person being called, to recognize the attempted fraud. However, this is often not the case, and the consequences can be not only financial but also psychological damage, as the person has become a victim of a crime. The attempt, now known as the "grandparent scam," to pose as a relative on the phone to elderly people, who is in need and needs a lot of money from grandma or grandpa very quickly, is also enjoying increasing popularity among organized gangs.

[0004] US 2020 / 0099781 A1 and US 2015 / 0281925 A1 disclose methods for identifying unwanted calls based on call pattern analysis. EP 3 443 723 B1, on the other hand, discloses such a method that can perform system-internal voice analysis.

[0005] In general, other calls, which may involve contracts of any kind, are also a popular means of defrauding people and sometimes extorting large sums of money.

[0006] The present invention is therefore based on the object of specifying techniques that enable connection subscribers to be efficiently protected from fraudulent calls.

[0007] The features of the various aspects of the invention or the various embodiments described below can be combined with one another unless this is explicitly excluded or technically mandatory.

[0008] According to the invention, a method is provided for identifying unwanted calls to a call-capable terminal of a subscriber in a communications network, the method comprising a voice analysis: Voice analysis is configured to extract speech as audio signals from the call information data and compare them with known voice profiles. This comparison is used to identify unwanted calls. Known voice profiles, such as those of relatives of the subscriber, can be stored in a database. If the voice analysis shows no match with the voices stored in the database, this indicates a potentially unwanted call. Another option is to store dialects as voice profiles. Typically, a dialect corresponding to the location of the subscriber's residence would be stored.As a rule, it can be assumed that the subscriber is mostly being called "intentionally" by people who live nearby and therefore have at least some echoes of the local dialect in their language; if the caller's dialect deviates too much from the local or national dialect, this may indicate an unwanted call.

[0009] The process thus enables automated detection of unwanted calls without the subscriber, i.e., the called party, having to actively recognize that the call is unwanted. This is particularly useful for efficiently protecting seniors from unwanted calls. The three analysis methods listed can be combined as desired. A combination of the three analysis methods enables even more accurate detection of unwanted calls.

[0010] The algorithm conveniently detects (i.e., identifies) an unwanted call when a defined probability threshold has been exceeded. This advantageously allows for adjusting the sensitivity of the unwanted call detection method and preventing too many calls from being marked as unwanted. In particular, it is possible to define a separate probability threshold for each of the three analysis methods and for each combination of the three analysis methods.

[0011] Preferably, technical countermeasures are initiated after the identification server has identified an unwanted call. One such countermeasure could be, for example, playing an audio signal to the subscriber during the call to warn them. Such an audio signal could, for example, say "Attention, this may be an unwanted call from a fraudster." Another possible countermeasure is to technically interrupt the call or, if necessary, even block the caller's connection. Preferably, such a blocking of the caller's connection is only carried out if the caller is repeatedly identified as an unwanted caller, especially within a predefined period of time. This offers the advantage that the subscriber can be effectively protected from the unwanted call using technical automation.In particular, the subscriber does not have to take any action himself.

[0012] Preferably, typical call patterns are generated by the identification server using machine learning. Machine learning algorithms in particular are very efficient at creating patterns, including typical call patterns. In particular, the calls can be divided into specific classes. A class can, for example, relate to the time of day of incoming calls to the subscriber. For example, a typical call pattern can have a class for each hour of the day and specify an average number of calls at these respective hours. If the call falls into a class in which the subscriber typically receives no calls at all, this may indicate an unwanted call. In particular, the typical call pattern can be regularly regenerated and / or updated.

[0013] Various event points can also be defined that trigger an update to the typical call pattern. For example, such an event can be triggered by a customer's contract change and / or when the subscriber is temporarily abroad. Both events can result in a change to the typical call pattern.

[0014] Whether the subscriber, in particular a subscriber to a mobile network, is located abroad can be automatically evaluated using the GPS data of their device, for example. If the subscriber is located abroad, the typical call pattern can also be adjusted so that the entries at which the calls typically take place are moved to a different class of the typical call pattern according to the time difference. This will be illustrated using a concrete example. If the typical call pattern has a class of the number of calls from 12-1 p.m., with typically 5 calls taking place within this time window, and if the subscriber is in a time zone with a time difference of 2 hours, the 5 calls can be moved to the class of calls from 2-3 p.m.

[0015] Preferably, deviations from the typical call pattern can be taken into account in the form of special event days and / or a call whitelist. For example, the date of the subscriber's birthday can be stored as information about the typical call pattern. Of course, other event days, such as holidays, can also be stored. On such days, especially birthdays, it can be expected that there will be deviations from the otherwise typical call pattern. For example, the detection of unwanted calls can be selectively deactivated on such days or the corresponding probability thresholds can be lowered. In addition, it can be taken into account whether the calls that call at atypical times, for example, come from a whitelist.The connection identifiers on this whitelist are calls that are known to be "desired" calls for the subscriber. Therefore, even if they would otherwise be marked as unwanted calls, they are still marked as "desired" calls. In any case, no technical countermeasures are initiated for calls marked as desirable calls.

[0016] In one embodiment of the invention, the typical call pattern is stored on the identification server. This offers the advantage that the identification server can efficiently execute the algorithm for detecting unwanted calls without generating additional data volume when transmitting the typical call patterns.

[0017] In one embodiment, the audio signals, particularly in the case of context analysis, are recorded by an external recording device. This enables the possibility of additional monitoring of the audio signals using an independent technical means. For example, a smart speaker with its microphone could record the conversation and send the call information data to the identification server. This can be particularly advantageous if the terminal device has been hacked by the fraudster.

[0018] According to a second aspect of the invention, a communication network is provided for identifying unwanted calls to a call-capable terminal of a subscriber in a communication network, the communication network comprising: • a call-capable terminal device of a subscriber; ◯ typically a telephone, especially a smartphone; • an identification server; ◯ with a computer unit for executing algorithms; • Communication connections between devices of the communication network, in particular between the call-capable terminal and the identification server; ◯ In addition, the call-capable terminal of the potential fraudster, hereinafter referred to as the additional call-capable terminal, may also be integrated into the communications network or communicate with the communications network via another communications network. The latter case occurs, for example, if the additional terminal is assigned to a different network service provider than the call-capable terminal of the subscriber. • wherein the identification server is configured to carry out the method of the method described above.

[0019] The communication network thus enables automated detection of unwanted calls without the subscriber, i.e. the called party, having to actively recognize that it is an unwanted call.

[0020] In one embodiment, the communications network has an interruption unit for selectively interrupting a communications connection. This offers the advantage that a communications connection associated with a potential fraudster can be specifically interrupted, effectively protecting the other subscribers from this fraud.

[0021] Further advantageous features of the present invention are defined in the patent claims.

[0022] In the following, preferred embodiments of the present invention are explained with reference to the accompanying figures: Fig. 1: shows an inventive communication network for detecting unwanted calls.

[0023] Numerous features of the present invention are explained in detail below using preferred embodiments. The present disclosure is not limited to the specifically mentioned feature combinations. Rather, the features mentioned here can be combined in any way to form embodiments of the invention, unless expressly excluded below.

[0024] Fig. 1 shows a communication network 10 for detecting unwanted calls.

[0025] The communications network 10 comprises a terminal 12 of a subscriber, wherein the terminal 12 is connected to an interface 16 of the communications network 10 via communications links 14. This interface 16 of the communications network 10 can, in particular, be formed by a switching center 16. An identification server 18 can also be provided in the switching center 16 or at least be in communication connection with the switching interface 16 via a communications link 14. The identification server 18 can also be provided such that it is included in a communications link 14 between the terminal 12 and the further terminal 20 of a potential fraudster. With such an arrangement, the identification server 18 can intercept call information data between the terminal 12 and the further terminal 20 particularly efficiently.

[0026] An algorithm for detecting unwanted calls is implemented on the identification server 18.

[0027] The identification server 18 receives call information data in step 24 and can evaluate it for an unwanted call. This call information data can be generated directly by a call request from the other terminal. However, call information data can also be transmitted, for example, from the terminal 12 to the identification server 18. Typically, in the latter case, step 22 will take place first: Step 22: Connecting the call from the additional terminal 20 to the subscriber via the communications network 10 to the subscriber's terminal 12. A caller typically dials the subscriber's number and is forwarded to the subscriber by the exchange 16 of the communications network 10. In particular, the identification server 18 can be located in the exchange 16 or can be in a communications connection 14 with this exchange 16. Connecting the call is, however, an optional first step, since unwanted calls can also be identified—as explained below—without necessarily connecting the call first.

[0028] After the call information data has been received in step 24, the algorithm passes it to either one or more identification methods. To do this, the algorithm uses at least the following three identification methods, which can be combined in any way: • Call pattern analysis 26: This involves analyzing and monitoring the telephony behavior of the called subscriber, or rather their terminal device 12. Using machine learning methods, the subscriber's typical calling behavior is learned over time and can be stored as a typical call pattern. Deviations from this typical call pattern, such as a high call volume within a short period of time, unusual call lengths, or outgoing calls directly after incoming calls, are identified as an anomaly, indicating a possible unwanted call. Appropriate technical countermeasures can be initiated if necessary. • Voice analysis 28: Using voice recognition algorithms, it can be determined whether the caller is a person known to the subscriber, for example, a family member or someone from their circle of acquaintances. This effectively helps identify fraudsters posing as family members. Appropriate technical countermeasures can be implemented if necessary. • Context analysis 30: Here, the conversation between the caller and the called party is first listened to and then transcribed in a further step. Suspicious words and terms can be extracted from the words spoken by both the caller and the called party, indicating an attempted fraud by the caller. If this is detected, both the called party and / or even their relatives can be automatically warned using technical countermeasures. Listening, in particular the subscriber's responses, can be carried out using a recording device 34, for example, a smart speaker 34. The smart speaker 34 can send the recorded audio signals as call information data to the identification server 18 via the communication link 14.The recording device 34 can also be configured as a device that is connected between the telephone socket and the subscriber's terminal device 12, thereby directly capturing the audio data. Another possibility is for an app on the user's terminal device 12 to extract the audio signals and then send them to the identification server 18 as call information data.

[0029] If the algorithm determines that the call is unwanted with a probability that is above a predefined threshold, appropriate countermeasures 32 can be initiated.

[0030] The invention according to the method thus makes it possible to effectively warn and protect subscribers from unwanted calls.

Claims

[1] Method for identifying unwanted calls to a call-capable terminal (12) of a subscriber in a communication network, comprising receiving audio signals of the call that are recorded by an external recording device and sent to an identification server (18) of the communication network (10), wherein an algorithm for identifying unwanted calls is implemented on the identification server, which algorithm is programmed to carry out a voice analysis wherein a dialect corresponding to the location of the subscriber is stored as a voice profile, wherein the voice analysis is set up to extract speech as audio signals from the call information data and to compare these with the known voice profiles and to identify an unwanted call by means of the comparison, wherein it indicates an unwanted call if the dialect of the caller deviates too much from the local or national dialect and After an unwanted call has been identified by the identification server, a technical countermeasure is initiated. [2] Method according to claim 1, characterized by that the algorithm detects an unwanted call when a defined probability threshold has been exceeded. [3] Method according to one of the preceding claims, characterized bythat as a technical countermeasure a warning signal is played to the subscriber or the call is interrupted. [4] A communication network for identifying unwanted calls to a call-capable terminal of a subscriber in a communication network (10), comprising: • a call-capable terminal (12) of a subscriber; • an external recording device suitable for recording audio signals of a call and sending them to an identification server (18); • Communication connections (14) between devices of the communication network, in particular between the call-capable terminal and the identification server; characterized by that the identification server is set up to carry out the method according to one of claims 1-3. [5] Communication network according to claim 4, characterized bythat the communication network has an interruption unit for interrupting a selective communication connection. [6] Identification server for identifying unwanted calls arranged to carry out the steps of the method according to one of claims 1-3.

Images