Call detection early warning method and device, equipment and storage medium

Abstract

The invention relates to the field of computers, and discloses a call detection early warning method and device, equipment and a storage medium. The method comprises the following steps: when a call is generated, acquiring call voice, extracting audio features in the call voice, and recognizing emotional features of a user during the call through a voice recognition technology according to the audio features; converting the call voice into a text through a voice recognition technology, and inputting the text into a pre-trained language model for keyword detection; according to the emotional characteristics, the detected keywords and a preset number filtering rule, judging whether the call number is an abnormal number or not by adopting a multi-layer filtering mechanism; and if yes, intercepting the current call through the API interface, and sending early warning information to the user. In the embodiment of the invention, more accurate call control interception can be realized.

Description

Technical Field

[0001] The present invention relates to the field of computers, and in particular to a call detection and early warning method, device, equipment and storage medium. Background Art

[0002] With the popularity of smart phones and mobile Internet, teenagers use mobile phones more and more frequently for calls and Internet activities in their daily lives. However, this also brings many network security risks. In order to protect teenagers from these risks, many institutions and organizations have taken different measures. For example, some mobile phone manufacturers have also launched mobile phone products specifically for teenagers. These products have the function of actively blocking harassing calls, text messages and bad websites, helping parents better manage their children's mobile phone usage.

[0003] However, existing application software can only intercept calls by identifying phone numbers, and it is difficult to customize call interception according to different application scenarios and user needs. It also lacks more fine-grained supervision and interception, making it difficult to achieve more precise control and interception. Summary of the invention

[0004] The main purpose of the present invention is to solve the technical problem of how to achieve more accurate call control and interception.

[0005] A first aspect of the present invention provides a call detection and early warning method, the call detection and early warning method comprising:

[0006] When a call occurs, the call voice is acquired, audio features in the call voice are extracted, and based on the audio features, the emotional features of the user during the call are identified through speech recognition technology;

[0007] The call speech is converted into text by the speech recognition technology, and the text is input into the pre-trained language model for keyword detection;

[0008] Based on the emotional characteristics, detected keywords and preset filtering number rules, a multi-layer filtering mechanism is used to determine whether the number being called is an abnormal number;

[0009] If so, the current call is intercepted through the API interface and a warning message is sent to the user.

[0010] Optionally, in a first implementation of the first aspect of the present invention, the determining whether the call number is an abnormal number by using a multi-layer filtering mechanism according to the emotional features, the detected keywords and the preset filtering number rules includes:

[0011] Collecting spoken language data in different scenarios, and establishing a corpus vocabulary based on the spoken language data;

[0012] According to the corpus word library, a synonym library is established by similarity calculation;

[0013] Determining whether to perform synonymous replacement on the keyword based on the synonym database and the context information of the keyword;

[0014] If so, a multi-layer filtering mechanism is used according to the emotional characteristics, the replaced keywords and the preset filtering number rules to determine whether the call number is an abnormal number.

[0015] Optionally, in a second implementation of the first aspect of the present invention, the determining whether the call number is an abnormal number by using a multi-layer filtering mechanism according to the emotional features, the replaced keywords and the preset filtering number rules includes:

[0016] Determine the part of speech of the keyword by part of speech tagging technology;

[0017] In combination with the context information, according to the part of speech of the keyword, the meaning of the keyword in the context is identified;

[0018] According to the meaning of the keyword in the context, the emotional characteristics and the preset number filtering rules, a multi-layer filtering mechanism is adopted to determine whether the call number is an abnormal number.

[0019] Optionally, in a third implementation of the first aspect of the present invention, the determining whether the call number is an abnormal number by using a multi-layer filtering mechanism according to the meaning of the keyword in the context, the emotional characteristics and the preset number filtering rules includes:

[0020] Obtaining a configuration parameter file, and defining the keyword and preset filtering number rules according to the configuration parameter file;

[0021] Through the plug-in mechanism, the application scenarios of the keywords and preset filtering number rules are expanded;

[0022] According to the contextual meaning of the keyword in the application scenario, the preset number filtering rules, and the emotional characteristics, a multi-layer filtering mechanism is adopted to determine whether the call number is an abnormal number.

[0023] Optionally, in a fourth implementation of the first aspect of the present invention, after determining whether the call number is an abnormal number by using a multi-layer filtering mechanism according to the contextual meaning of the keyword in the application scenario, the preset number filtering rule, and the emotional feature, the method further includes:

[0024] Sending the abnormal number to the cloud platform, so that before acquiring the call voice next time, the cloud platform is used to determine whether the call number is the abnormal number;

[0025] If so, the step of obtaining the call voice and extracting the audio features in the call voice is entered.

[0026] Optionally, in a fifth implementation of the first aspect of the present invention, after determining whether the call number is an abnormal number by using a multi-layer filtering mechanism according to the emotional features, the detected keywords and the preset filtering number rules, the method further includes:

[0027] Regularly screen the keywords, remove irrelevant keywords, and use synonyms, antonyms and related words to expand the keyword library;

[0028] An incremental pre-training strategy is adopted to periodically select and adjust parameters in the pre-trained language model to update the pre-trained language model.

[0029] Optionally, in a sixth implementation of the first aspect of the present invention, before intercepting the current call through the API interface and sending the warning information to the user, the method further includes:

[0030] Get the incoming call time of the current call and the corresponding user hang-up behavior;

[0031] According to the incoming call time and the user's hang-up behavior, judging whether the current call is a harassing call by using a harassing call recognition model;

[0032] If so, the process proceeds to the step of intercepting the current call through the API interface and sending an early warning message to the user.

[0033] The second aspect of the present invention provides a call detection and warning device, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via lines; the at least one processor calls the instructions in the memory so that the call detection and warning device executes the above-mentioned call detection and warning method.

[0034] A third aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned call detection and early warning method.

[0035] In an embodiment of the present invention, when a call is generated, the call voice is obtained, the audio features in the call voice are extracted, and based on the audio features, the emotional features of the user during the call are identified through voice recognition technology; the call voice is converted into text through the voice recognition technology, and the text is input into a pre-trained language model for keyword detection; based on the emotional features, the detected keywords and the preset filtering number rules, a multi-layer filtering mechanism is used to determine whether the call number is an abnormal number; if so, the current call is intercepted through the API interface, and an early warning message is sent to the user. In the present invention, by pre-setting the filtering number rules, users can set specific contacts or keywords according to their needs to realize a personalized early warning function, and through emotional feature recognition and keyword detection, calls can be monitored and intercepted according to the emotional state of the caller and the content of the call, thereby realizing more accurate call control and interception. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A schematic diagram of an embodiment of a call detection and early warning method in an embodiment of the present invention;

[0037] Figure 2 A schematic diagram of an embodiment of a call detection and early warning device in an embodiment of the present invention;

[0038] Figure 3 The figure is a schematic diagram of an embodiment of a call detection and early warning device in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] Embodiments of the present invention provide a call detection and early warning method, device, equipment and storage medium.

[0040] The embodiments disclosed in the present invention will be described in more detail below with reference to the accompanying drawings. Although certain embodiments disclosed in the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments disclosed in the present invention are only for exemplary purposes and are not intended to limit the scope of protection disclosed in the present invention.

[0041] In the description of the embodiments disclosed in the present invention, the term "including" and similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0042] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the call detection and early warning method in the embodiment of the present invention includes:

[0043] S100, when a call occurs, acquire the call voice, extract the audio features in the call voice, and identify the emotional features of the user during the call through voice recognition technology based on the audio features.

[0044] In this embodiment, a real-time message stream processing framework (such as Apache Kafka, Apache Flink) is used to process the data stream in real time. When the call is connected, the real-time collection of the call voice is authorized, and the collected call voice is preprocessed, including noise elimination and frame windowing, to improve the accuracy of subsequent feature extraction. Then, the audio features in the preprocessed call voice are extracted. Optionally, the Librosa library can be used for audio feature extraction, and the high-pass filter is used to perform spectrum shaping on the voice signal to reduce low-frequency noise and reduce the error rate. Monitor the fluctuations in the channel, such as trembling sounds, sudden sharp channels, etc. These fluctuations may reflect abnormal emotions, extract the sounds and call objects with abnormal emotional fluctuations, and as a record, the other party can be temporarily identified as a threatening party, which will be occasionally interrupted and prompted during the next call process. If the user acquiesces to the behavior, it will be completely blocked (the incoming and outgoing calls are directly hung up). By analyzing the fluctuations in the channel, the speaker's pitch and intensity changes can be analyzed, and the fluctuations of emotions can be identified. For example, in an angry mood, the speaker's pitch may increase, and in a nervous state, the volume may increase.

[0045] S200, through speech recognition technology, converts the call speech into text and inputs the text into a pre-trained language model for keyword detection.

[0046] In this embodiment, first, the voice data needs to be preprocessed, including noise reduction, segmentation, feature extraction, etc., and the preprocessed call voice is converted into text. This step is usually completed using automatic speech recognition technology. Then, the converted text is preprocessed, including word segmentation, stop word removal, stem extraction, etc., so that the text is more suitable for input into the pre-trained language model. Finally, the pre-processed text is input into the pre-trained language model for real-time keyword detection, wherein the pre-trained language model adopts the BERT or RoBERTa model. Both BERT and RoBERTa are models based on the Transformer architecture, which can capture the contextual relationship in the text and are very suitable for keyword detection tasks. By using pre-trained language models (such as BERT, RoBERTa) to identify keywords in call voice, not only can the accuracy of recognition be improved, but also the dependence on a large amount of annotated data can be reduced, thereby reducing development costs and time.

[0047] S300, based on the emotional features, detected keywords and preset filtering number rules, a multi-layer filtering mechanism is used to determine whether the calling number is an abnormal number.

[0048] In this embodiment, the preset filtering number rules include user-defined phone number rules and channel fluctuation ranges, wherein the phone number rules include area code rules and number rules, and the channel fluctuation range refers to the channel fluctuation range allowed by the user during a call. The use of a multi-layer filtering mechanism to screen the call numbers layer by layer can effectively improve the call quality and security. Specifically, first, after the call number is filtered by the national anti-fraud system, it is filtered through the user-defined preset filtering number rules, and then the call number is filtered again through the shared user big database, and then the call number is filtered again based on the abnormal emotional fluctuations analyzed according to the user's channel information, and combined with context analysis to reduce false positives.

[0049] S400: If yes, intercept the current call through the API interface and send a warning message to the user.

[0050] In this embodiment, if the number of the current call is an abnormal number, the current call is intercepted through the Android API interface, the abnormal number is reported to the monitoring record, and an early warning message is sent to the user. The call detection and early warning function of the present invention can be integrated with other third-party application software through the API interface. The API integration allows third-party developers to build applications integrated with their services by providing a standardized and secure way to access and share data, thereby expanding the coverage of the call detection and early warning function. At the same time, the SDK is embedded in commonly used application software to ensure comprehensive coverage.

[0051] In an optional implementation of the first aspect of the present invention, a multi-layer filtering mechanism is used according to the emotional characteristics, the detected keywords and the preset filtering number rules to determine whether the call number is an abnormal number, including:

[0052] Collect conversation corpus in different scenarios, and build a corpus vocabulary based on the conversation corpus; build a synonym database based on the corpus vocabulary by similarity calculation; based on the synonym database, combined with the contextual information of the keywords, determine whether to perform synonymous replacement on the keywords; if so, use a multi-layer filtering mechanism based on emotional characteristics, replaced keywords and preset filtering number rules to determine whether the call number is an abnormal number.

[0053] In this embodiment, first, a large amount of spoken language corpus needs to be collected. These spoken language corpus should cover various usage scenarios of the target language, including formal text, informal text, dialogue, etc. Then, the collected spoken language corpus is segmented, the sentences are decomposed into words or phrases, and the results after segmentation are cleaned to remove invalid words, stop words, etc., to ensure the purity of the synonym library. At the same time, the vocabulary is standardized, such as unifying uppercase and lowercase, removing special characters, etc., to establish a vocabulary. Secondly, the similarity of the vocabulary in the vocabulary is calculated, and algorithms such as cosine similarity and edit distance are usually used to measure the similarity between vocabulary. Vocabularies with high similarity are regarded as synonyms. According to the results of the similarity calculation, a synonym library is constructed, which contains each vocabulary and its synonym list. According to the synonym library, combined with the context information of the keyword, it is determined whether to perform synonymous replacement on the keyword, that is, whether there are homophones or variant characters. If the keyword needs to be replaced synonymously, all synonyms of the keyword are searched in the synonym library, and the most appropriate vocabulary is selected for replacement according to the context.

[0054] In an optional implementation of the first aspect of the present invention, a multi-layer filtering mechanism is used according to emotional features, replaced keywords and preset filtering number rules to determine whether the call number is an abnormal number, including:

[0055] The part of speech of the keyword is determined through part-of-speech tagging technology; the meaning of the keyword in the context is identified based on the part of speech of the keyword in combination with context information; based on the meaning of the keyword in the context, emotional characteristics and preset filtering number rules, a multi-layer filtering mechanism is used to determine whether the call number is an abnormal number.

[0056] In this embodiment, after replacing the keywords, first, the part of speech of each keyword is determined using part-of-speech tagging technology, which helps to understand the function and role of words in sentences, and select the correct word meaning through semantic analysis, for example, distinguishing the meaning of "bank" in different contexts, such as verbs and nouns. Then, through context parsing, according to the part of speech of the keyword, the meaning of the keyword in the context is identified. Specifically, semantic analysis technology is used to capture the co-occurrence relationship between the keyword and other words, so as to understand the meaning of the keyword, which includes using word embedding methods such as Word2Vec or GloVe to capture the semantic meaning of keywords and sentences. Finally, according to the meaning of the keyword in the context, the fluctuation in the sound channel and the preset filtering number rules, a multi-layer filtering mechanism is used to determine whether the number being called is an abnormal number.

[0057] In an optional implementation of the first aspect of the present invention, based on the meaning of the keyword in the context, the emotional characteristics and the preset number filtering rules, a multi-layer filtering mechanism is used to determine whether the call number is an abnormal number, including:

[0058] Obtain the configuration parameter file, and define keywords and preset filtering number rules according to the configuration parameter file; expand the application scenarios of keywords and preset filtering number rules through the plug-in mechanism; use a multi-layer filtering mechanism based on the contextual meaning of keywords in the application scenario, preset filtering number rules, and emotional characteristics to determine whether the calling number is an abnormal number.

[0059] In this embodiment, a configurable parameter file is provided to allow users to customize keywords and preset filtering number rules. The call interception application adopts Google's component mode, provides a main program and conventional rules, and defines the writing format of the rules, and users can spontaneously customize the rules. At the same time, a plug-in mechanism is developed, which allows third parties to extend functions to meet the needs of different scenarios, so as to expand the application scenarios of keywords and preset filtering number rules. Then, according to the contextual meaning of keywords in different application scenarios, preset filtering number rules, and emotional characteristics, it is judged whether the number of the call is an abnormal number.

[0060] In an optional implementation of the first aspect of the present invention, according to the contextual meaning of the keywords in the application scenario, the preset number filtering rules, and the emotional characteristics, a multi-layer filtering mechanism is adopted to determine whether the call number is an abnormal number, and further includes:

[0061] The abnormal number is sent to the cloud platform, so that before the next call voice is obtained, the cloud platform can be used to determine whether the call number is an abnormal number; if so, the step of obtaining the call voice and extracting the audio features in the call voice is entered.

[0062] In this embodiment, when the call number is an abnormal number, the abnormal number is sent to the cloud platform, and the cloud platform can use big data analysis and machine learning algorithms to further verify and classify these numbers. The call detection and early warning function in the present invention is applicable to SDKs of different platforms (such as iOS, Android, Windows, macOS), ensuring cross-platform compatibility, and realizing cross-platform data synchronization and processing through cloud services to ensure data consistency and real-time. Before acquiring the call voice next time, based on the abnormal number information previously stored by the cloud platform, it is judged in real time whether the number of the current call is an abnormal number. If an abnormal number is identified, corresponding measures can be taken, such as interception or marking.

[0063] In an optional implementation of the first aspect of the present invention, after determining whether the number being called is an abnormal number by using a multi-layer filtering mechanism based on the emotional features, the detected keywords and the preset filtering number rules, the method further includes:

[0064] Regularly screen keywords, remove irrelevant keywords, and use synonyms, antonyms and related words to expand the keyword library; adopt an incremental pre-training strategy, and regularly adjust the parameters in the pre-trained language model to update the pre-trained language model.

[0065] In this embodiment, the keyword library is regularly screened to remove irrelevant and low-quality keywords, retain high-quality and highly relevant keywords, and regularly check and update the keyword list, track keywords, use synonyms, synonyms and related words to expand the keyword library, and specifically, use TF and TFIDF technology to process the data set to find new keywords. An incremental pre-training strategy is adopted to update only the parameters related to a specific task, avoiding equal optimization of all parameters to update the pre-trained language model. At the same time, a user feedback mechanism is established to collect false positives and continuously optimize the model.

[0066] In an optional implementation of the first aspect of the present invention, before intercepting the current call through the API interface and sending the warning information to the user, the method further includes:

[0067] Get the incoming call time of the current call and the corresponding user hang-up behavior; based on the incoming call time and the user hang-up behavior, use the spam call recognition model to determine whether the current call is a spam call; if so, proceed to the step of intercepting the current call through the API interface and sending a warning message to the user.

[0068] In this embodiment, a large amount of call record data needs to be collected, including the start time, end time, connection status (whether to hang up), waiting time, etc. of the call. These data can be obtained from the operator's database or collected through a user feedback system. The collected data is analyzed to extract features related to harassing calls, such as frequent slow connection or hang-up behaviors. Machine learning algorithms such as random forests, Bayesian networks, etc. are used to train the extracted features and establish a harassing call recognition model. This model can learn the difference between normal calls and harassing calls. After the model training is completed, it needs to be evaluated to ensure its accuracy and reliability. When suspicious call behavior is detected through the harassing call recognition model based on the incoming call time and the user's hang-up behavior, the system can automatically mark it as a harassing call and take corresponding interception measures.

[0069] See also Figure 2 According to a second aspect of the present invention, a call detection and early warning device is provided, and the call detection and early warning device comprises:

[0070] The emotion feature recognition module 10 is used to obtain the call voice when a call is made, extract the audio features in the call voice, and recognize the emotion features of the user during the call through the voice recognition technology based on the audio features;

[0071] The keyword detection module 20 is used to convert the call speech into text through speech recognition technology, and input the text into the pre-trained language model for keyword detection;

[0072] The abnormal number identification module 30 is used to determine whether the number being called is an abnormal number by using a multi-layer filtering mechanism based on emotional characteristics, detected keywords and preset filtering number rules;

[0073] The warning information prompt module 40 is used to intercept the current call through the API interface and send a warning message to the user.

[0074] In an optional implementation of the second aspect of the present invention, the abnormal number identification module 30 is also used to collect conversation corpus in different scenarios, and establish a corpus vocabulary based on the conversation corpus; establish a synonym database based on the corpus vocabulary by similarity calculation; based on the synonym database, combined with the context information of the keywords, determine whether to perform synonymous replacement on the keywords; if so, use a multi-layer filtering mechanism based on emotional characteristics, replaced keywords and preset filtering number rules to determine whether the call number is an abnormal number.

[0075] In an optional implementation of the second aspect of the present invention, the abnormal number identification module 30 is also used to determine the part of speech of the keyword through part-of-speech tagging technology; in combination with context information, identify the meaning of the keyword in the context according to the part of speech of the keyword; based on the meaning of the keyword in the context, emotional characteristics and preset filtering number rules, a multi-layer filtering mechanism is used to determine whether the call number is an abnormal number.

[0076] In an optional implementation of the second aspect of the present invention, the abnormal number identification module 30 is also used to obtain a configuration parameter file, and define keywords and preset filtering number rules based on the configuration parameter file; expand the application scenarios of keywords and preset filtering number rules through a plug-in mechanism; and use a multi-layer filtering mechanism based on the contextual meaning of keywords in the application scenario, preset filtering number rules, and emotional characteristics to determine whether the call number is an abnormal number.

[0077] In an optional implementation of the second aspect of the present invention, the call detection and warning device also includes a cloud platform data judgment module, which is used to send the abnormal number to the cloud platform, so that before obtaining the call voice next time, the cloud platform can be used to determine whether the call number is an abnormal number; if so, the step of obtaining the call voice and extracting audio features in the call voice is entered.

[0078] In an optional implementation of the second aspect of the present invention, the call detection and warning device also includes a data update module, which is used to regularly screen keywords, remove irrelevant keywords, and use synonyms, antonyms and related vocabulary to expand the keyword library; adopt an incremental pre-training strategy to regularly select and adjust the parameters in the pre-trained language model to update the pre-trained language model.

[0079] In an optional implementation of the second aspect of the present invention, the call detection and warning device also includes a harassing call judgment module, which is used to obtain the incoming call time of the current call and the corresponding user hang-up behavior; based on the incoming call time and the user hang-up behavior, through the harassing call recognition model, it is judged whether the current call is a harassing call; if so, it enters the step of intercepting the current call through the API interface and sending a warning message to the user.

[0080] Figure 3 1 is a schematic diagram of the structure of a call detection and warning device provided by an embodiment of the present invention. The call detection and warning device 500 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 510 (for example, one or more processors) and a memory 520, and one or more storage media 530 (for example, one or more mass storage devices) storing application programs 533 or data 532. Among them, the memory 520 and the storage medium 530 can be short-term storage or permanent storage. The program stored in the storage medium 530 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the call detection and warning device 500. Furthermore, the processor 510 can be configured to communicate with the storage medium 530 to execute a series of instruction operations in the storage medium 530 on the call detection and warning device 500.

[0081] The call detection based warning device 500 may also include one or more power supplies 540, one or more wired or wireless network interfaces 550, one or more input and output interfaces 560, and / or one or more operating systems 531, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that Figure 3 The illustrated call detection warning device structure does not constitute a limitation on the call detection-based warning device, and may include more or fewer components than illustrated, or a combination of certain components, or a different arrangement of components.

[0082] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the call detection and warning method.

[0083] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0084] In addition, although each operation is described in a specific order, this should be understood as requiring such operation to be performed in the specific order shown or in a sequential order, or requiring that all illustrated operations should be performed to obtain desired results. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single implementation in combination. On the contrary, the various features described in the context of a single implementation can also be implemented in multiple implementations individually or in any suitable sub-combination mode.

[0085] Although the subject matter has been described in language specific to structural features and / or methodological logical actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims.

Claims

1. A call detection and early warning method, characterized in that: The call detection and early warning method comprises: When a call occurs, the call voice is acquired, audio features in the call voice are extracted, and based on the audio features, the emotional features of the user during the call are identified through speech recognition technology; The call speech is converted into text by the speech recognition technology, and the text is input into the pre-trained language model for keyword detection; Based on the emotional characteristics, detected keywords and preset filtering number rules, a multi-layer filtering mechanism is used to determine whether the number being called is an abnormal number; If so, the current call is intercepted through the API interface and a warning message is sent to the user.

2. The call detection and early warning method according to claim 1, characterized in that: The method of using a multi-layer filtering mechanism based on the emotional characteristics, detected keywords and preset filtering number rules to determine whether the call number is an abnormal number includes: Collecting spoken language data in different scenarios, and establishing a corpus vocabulary based on the spoken language data; According to the corpus word library, a synonym library is established by similarity calculation; Determining whether to perform synonymous replacement on the keyword based on the synonym database and the context information of the keyword; If so, a multi-layer filtering mechanism is used according to the emotional characteristics, the replaced keywords and the preset filtering number rules to determine whether the call number is an abnormal number.

3. The call detection and early warning method according to claim 2, characterized in that: The method of using a multi-layer filtering mechanism based on the emotional characteristics, the replaced keywords and the preset filtering number rules to determine whether the call number is an abnormal number includes: Determine the part of speech of the keyword by part of speech tagging technology; In combination with the context information, according to the part of speech of the keyword, the meaning of the keyword in the context is identified; According to the meaning of the keyword in the context, the emotional characteristics and the preset number filtering rules, a multi-layer filtering mechanism is adopted to determine whether the call number is an abnormal number.

4. The call detection and early warning method according to claim 3, characterized in that: The method of using a multi-layer filtering mechanism to determine whether the call number is an abnormal number based on the meaning of the keyword in the context, the emotional characteristics and the preset number filtering rules includes: Obtaining a configuration parameter file, and defining the keyword and preset filtering number rules according to the configuration parameter file; Through the plug-in mechanism, the application scenarios of the keywords and preset filtering number rules are expanded; According to the contextual meaning of the keyword in the application scenario, the preset number filtering rules, and the emotional characteristics, a multi-layer filtering mechanism is adopted to determine whether the call number is an abnormal number.

5. The call detection and early warning method according to claim 4, characterized in that: After the method uses a multi-layer filtering mechanism to determine whether the call number is an abnormal number based on the contextual meaning of the keyword in the application scenario, the preset number filtering rule, and the emotional characteristics, the method further includes: The abnormal number is sent to the cloud platform, so that before the next call voice is obtained, the cloud platform is used to determine whether the call number is the abnormal number; If so, the step of obtaining the call voice and extracting the audio features in the call voice is entered.

6. The call detection and early warning method according to claim 1, characterized in that: After the multi-layer filtering mechanism is used to determine whether the call number is an abnormal number based on the emotional characteristics, the detected keywords and the preset filtering number rules, the method further includes: Regularly screen the keywords, remove irrelevant keywords, and use synonyms, antonyms and related words to expand the keyword library; An incremental pre-training strategy is adopted to periodically select and adjust parameters in the pre-trained language model to update the pre-trained language model.

7. The call detection and early warning method according to claim 1, characterized in that: Before intercepting the current call through the API interface and sending the warning information to the user, the method further includes: Get the incoming call time of the current call and the corresponding user hang-up behavior; According to the incoming call time and the user's hang-up behavior, judging whether the current call is a harassing call through a harassing call recognition model; If so, the process proceeds to the step of intercepting the current call through the API interface and sending an early warning message to the user.

8. A call detection and warning device, characterized in that: The call detection and early warning device comprises: The emotion feature recognition module is used to obtain the call voice when a call is made, extract the audio features in the call voice, and recognize the emotion features of the user during the call through the speech recognition technology based on the audio features; A keyword detection module, used to convert the call speech into text through the speech recognition technology, and input the text into a pre-trained language model for keyword detection; An abnormal number identification module is used to determine whether the number being called is an abnormal number by using a multi-layer filtering mechanism based on the emotional characteristics, detected keywords and preset filtering number rules; The warning information prompt module is used to intercept the current call through the API interface and send a warning message to the user.

9. A call detection and warning device, characterized in that: The call detection warning device comprises: a memory and at least one processor, the memory stores instructions, and the memory and the at least one processor are interconnected via a line; The at least one processor calls the instruction in the memory so that the call detection and warning device executes the call detection and warning method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the call detection and early warning method according to any one of claims 1 to 7 is implemented.

Images