Junk call identification method and device, computer equipment and storage medium

Abstract

The invention belongs to the technical field of artificial intelligence, and relates to a junk call identification method and device, computer equipment and a storage medium, and the method comprises the steps: obtaining initial audio data corresponding to a target call when the target call is accessed; preprocessing the initial audio data to obtain specified audio data; performing feature extraction on the specified audio data based on a pre-training language model to obtain a specified audio feature vector; calling a junk call recognition model; performing identification processing on the specified audio feature vector based on a junk call identification model to obtain a probability that the target call belongs to each preset junk call category; and screening out a target probability with the highest value from all probabilities, and generating a junk call recognition result corresponding to the target call based on the target probability. In addition, the invention also relates to a block chain technology, and the junk call recognition result can be stored in a block chain. Based on the use of the junk call recognition model, the junk call recognition efficiency and recognition accuracy are improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence development technology and financial technology, and in particular to a method, device, computer equipment and storage medium for identifying spam calls. Background Art

[0002] In the current telecommunications security field, with the prevalence of telecommunications fraud and spam calls, financial companies are facing unprecedented challenges. In order to deal with this problem, many companies have adopted spam call identification systems based on traditional audio fingerprint technology. However, this traditional method has exposed many shortcomings when processing large-scale data.

[0003] Traditional audio fingerprint technology mainly relies on feature extraction and comparison of audio signals. In spam call identification systems, this technology is often used to build a template database of fraudulent audio, and then compare the audio of the current call with these templates to identify potential fraudulent calls. However, when processing large-scale data, this method significantly increases its computational complexity and memory usage, resulting in low processing efficiency. In addition, due to the influence of computational complexity and memory usage, the system may not be able to complete the audio fingerprint extraction and comparison process in a short period of time, and thus cannot identify and intercept potential fraudulent calls in a timely manner. At the same time, due to the diversity and complexity of audio signals, traditional audio fingerprint technology may not be able to accurately identify and match all fraudulent audio, resulting in a decrease in the accuracy of the system. Summary of the invention

[0004] The purpose of the embodiments of the present application is to propose a method, apparatus, computer device and storage medium for identifying spam calls, so as to solve the technical problems that the existing spam call identification system based on traditional audio fingerprint technology has low processing efficiency and low accuracy in the process of spam call identification.

[0005] In order to solve the above technical problems, the embodiment of the present application provides a method for identifying spam calls, which adopts the following technical solution:

[0006] When a target phone is accessed, initial audio data corresponding to the target phone is obtained;

[0007] Preprocessing the initial audio data to obtain corresponding specified audio data;

[0008] Perform feature extraction on the specified audio data based on the pre-trained language model to obtain a corresponding specified audio feature vector;

[0009] Call the preset spam call identification model;

[0010] Performing recognition processing on the designated audio feature vector based on the spam call recognition model to obtain the probability that the target phone belongs to each preset spam call category;

[0011] A target probability with the highest value is screened out from all the probabilities, and a spam call identification result corresponding to the target phone is generated based on the target probability.

[0012] Furthermore, before the step of calling the preset spam call identification model, the method further includes:

[0013] Get a pre-built, labeled spam call dataset;

[0014] Dividing the spam call dataset into a training set and a validation set;

[0015] Calling a preset deep learning model and determining a target loss function and a target optimization algorithm corresponding to the deep learning model;

[0016] Based on the target loss function and the target optimization algorithm, the deep learning model is trained using the training set to obtain a corresponding first deep learning model;

[0017] Performing model adjustment on the first deep learning model based on the validation set to obtain a second deep learning model that meets preset performance requirements;

[0018] The second deep learning model is used as the spam call identification model.

[0019] Furthermore, the step of obtaining a pre-constructed labeled spam call dataset specifically includes:

[0020] Preprocessing the pre-collected voice call data to obtain audio data, and extracting features from the audio data to obtain an audio feature vector;

[0021] Performing clustering processing on the audio feature vector based on a preset target clustering algorithm to obtain a corresponding clustering result;

[0022] Performing an abnormality analysis on the clustering result to obtain a corresponding abnormality analysis result;

[0023] Based on the clustering result and the abnormal analysis result, filtering out misdialed call audio and spam call audio from the audio data;

[0024] Transcribing the spam call audio to obtain the corresponding spam call text;

[0025] Performing language recognition on the spam call text based on a preset language recognition model to obtain a corresponding language recognition result;

[0026] Based on the preset target language requirement and the language recognition result, filtering out the designated spam call audio in the non-target language from the spam call audio to obtain the filtered target spam call audio;

[0027] The target spam call audio is annotated based on a preset annotation strategy to obtain the spam call data set.

[0028] Furthermore, the step of preprocessing the initial audio data to obtain the corresponding specified audio data specifically includes:

[0029] Performing denoising processing on the audio data to obtain corresponding first audio data;

[0030] Performing voice activity detection processing on the first audio data to obtain corresponding second audio data;

[0031] The second audio data is used as the designated audio data.

[0032] Furthermore, the step of extracting features from the specified audio data based on the pre-trained language model to obtain a corresponding specified audio feature vector specifically includes:

[0033] Calling the pre-trained speech model;

[0034] Performing forward propagation processing on the specified audio data based on the pre-trained language model to obtain a corresponding hidden layer state;

[0035] Performing a mean pooling operation on the hidden layer state to obtain a corresponding feature vector;

[0036] The specified audio feature vector is generated based on the feature vector.

[0037] Furthermore, the step of generating a spam call identification result corresponding to the target phone based on the target probability specifically includes:

[0038] Obtaining a preset probability threshold;

[0039] Determining whether the target probability is greater than the probability threshold;

[0040] If yes, generating a spam call identification result indicating that the target phone is a spam call;

[0041] If not, a spam call identification result is generated indicating that the target phone is a normal phone.

[0042] Further, after the step of generating a spam call identification result corresponding to the target phone based on the target probability, the method further includes:

[0043] If the content of the spam call identification result is that the target phone is a spam call, obtaining a target spam call category corresponding to the target probability;

[0044] Generate a call analysis result corresponding to the target call based on the target spam call category and the spam call identification result;

[0045] The telephone analysis result is outputted.

[0046] In order to solve the above technical problems, the embodiment of the present application further provides a device for identifying spam calls, which adopts the following technical solution:

[0047] A first acquisition module, used for acquiring initial audio data corresponding to the target phone when the target phone is accessed;

[0048] A preprocessing module, used for preprocessing the initial audio data to obtain corresponding specified audio data;

[0049] An extraction module, used to extract features from the specified audio data based on a pre-trained language model to obtain a corresponding specified audio feature vector;

[0050] A calling module, used to call a preset spam call identification model;

[0051] An identification module, configured to perform identification processing on the specified audio feature vector based on the spam call identification model to obtain the probability that the target phone belongs to each preset spam call category;

[0052] The first generating module is used to select a target probability with the highest value from all the probabilities, and generate a spam call identification result corresponding to the target phone based on the target probability.

[0053] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0054] When a target phone is accessed, initial audio data corresponding to the target phone is obtained;

[0055] Preprocessing the initial audio data to obtain corresponding specified audio data;

[0056] Perform feature extraction on the specified audio data based on the pre-trained language model to obtain a corresponding specified audio feature vector;

[0057] Call the preset spam call identification model;

[0058] Performing recognition processing on the designated audio feature vector based on the spam call recognition model to obtain the probability that the target phone belongs to each preset spam call category;

[0059] A target probability with the highest value is screened out from all the probabilities, and a spam call identification result corresponding to the target phone is generated based on the target probability.

[0060] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0061] When a target phone is accessed, initial audio data corresponding to the target phone is obtained;

[0062] Preprocessing the initial audio data to obtain corresponding specified audio data;

[0063] Perform feature extraction on the specified audio data based on the pre-trained language model to obtain a corresponding specified audio feature vector;

[0064] Call the preset spam call identification model;

[0065] Performing recognition processing on the designated audio feature vector based on the spam call recognition model to obtain the probability that the target phone belongs to each preset spam call category;

[0066] A target probability with the highest value is screened out from all the probabilities, and a spam call identification result corresponding to the target phone is generated based on the target probability.

[0067] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0068] When the target phone is connected to the present application, the initial audio data corresponding to the target phone is obtained; the initial audio data is then preprocessed to obtain the corresponding designated audio data; the designated audio data is then feature extracted based on the pre-trained language model to obtain the corresponding designated audio feature vector; the preset spam call recognition model is subsequently called; the designated audio feature vector is further identified and processed based on the spam call recognition model to obtain the probability that the target phone belongs to each preset spam call category; finally, the target probability with the highest value is screened out from all the probabilities, and a spam call recognition result corresponding to the target phone is generated based on the target probability. When the target phone is connected to the present application, the initial audio data of the target phone is obtained, and the initial audio data is preprocessed to obtain the designated audio data, and then the designated audio data is feature extracted based on the use of the pre-trained language model to obtain the designated audio feature vector, and then the designated audio feature vector is identified and processed based on the use of the spam call recognition model, so that the efficient identification and accurate classification of spam calls can be automatically and accurately completed, thereby improving the recognition efficiency and recognition accuracy of spam calls. Compared with traditional audio fingerprint technology, this application has lower computational complexity and memory usage when processing large-scale data, and can adapt to the rapid changes in spam call patterns, thereby effectively improving the real-time and accuracy of spam call identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0070] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0071] Figure 2 A flowchart of an embodiment of a method for identifying spam calls according to the present application;

[0072] Figure 3 is a structural schematic diagram of an embodiment of a junk call identification device according to the present application;

[0073] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0074] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0075] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0076] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0077] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0078] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0079] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to the laptop computer 1011, tablet computer 1012 or mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0080] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0081] It should be noted that the method for identifying spam calls provided in the embodiments of the present application is generally executed by a server / terminal device, and accordingly, the apparatus for identifying spam calls is generally disposed in the server / terminal device.

[0082] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0083] Continue to refer Figure 2 , shows a flow chart of an embodiment of a method for identifying spam calls according to the present application. According to different requirements, the order of the steps in the flow chart can be changed, and some steps can be omitted. The method for identifying spam calls provided in the embodiment of the present application can be applied to any scenario where spam call identification is required, and the method for identifying spam calls can be applied to products in these scenarios, for example, spam call identification in the financial insurance field. The method for identifying spam calls includes the following steps:

[0084] Step S201, when a target phone is connected, initial audio data corresponding to the target phone is obtained.

[0085] In this embodiment, the method for identifying spam calls is executed on an electronic device (eg, Figure 1The server / terminal device shown in the figure) can obtain the initial audio data of the target phone through a wired connection or a wireless connection. It should be noted that the above-mentioned wireless connection methods may include but are not limited to 3G / 4G / 5G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or developed in the future. The executor of the present application may be a spam call identification system, or simply referred to as a system. The system is a telephone network. When a new target phone is connected, the spam call identification system will collect the initial audio data of the target phone in real time. The present application can realize efficient identification and accurate classification of spam calls, improve the real-time and accuracy of spam call processing, and reduce the consumption of computing resources. The present application is particularly suitable for business fields such as insurance, banking and asset management that need to process a large amount of telephone data, and can help financial companies improve their defense capabilities against spam calls and protect customers from telephone fraud and harassment.

[0086] Step S202: pre-process the initial audio data to obtain corresponding designated audio data.

[0087] In this embodiment, the specific implementation process of preprocessing the initial audio data to obtain the corresponding specified audio data will be further described in detail in subsequent specific embodiments of the present application, and will not be elaborated on here.

[0088] Step S203: extract features from the specified audio data based on the pre-trained language model to obtain a corresponding specified audio feature vector.

[0089] In this embodiment, the specific implementation process of extracting features from the specified audio data based on the pre-trained language model to obtain the corresponding specified audio feature vector will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated on here.

[0090] Step S204, calling a preset spam call identification model.

[0091] In this embodiment, the model construction process of the above spam call identification model will be further described in detail in subsequent specific embodiments of the present application, and will not be elaborated in detail here.

[0092] Step S205 , performing recognition processing on the designated audio feature vector based on the spam call recognition model to obtain the probability that the target phone belongs to each preset spam call category.

[0093] In this embodiment, the extracted specified audio feature vector can be transmitted to the deployed spam call identification model through an interface. After receiving the above specified audio feature vector, the spam call identification model will perform forward propagation calculation and output the corresponding classification result. The classification result is usually a probability distribution, indicating the probability that the target phone belongs to each preset spam call category. Among them, the spam call categories may include fraud, sales promotion, and harassment.

[0094] Step S206, selecting a target probability with the highest value from all the probabilities, and generating a spam call identification result corresponding to the target phone based on the target probability.

[0095] In this embodiment, the probability that all target phones belong to each spam call category can be numerically compared, and the probability with the highest value can be selected as the target probability. In addition, the specific implementation process of generating the spam call identification result corresponding to the target phone based on the target probability will be further described in detail in the subsequent specific embodiments of this application, and will not be elaborated on here.

[0096] When the target phone is connected to the present application, the initial audio data corresponding to the target phone is obtained; the initial audio data is then preprocessed to obtain the corresponding designated audio data; the designated audio data is then feature extracted based on the pre-trained language model to obtain the corresponding designated audio feature vector; the preset spam call recognition model is subsequently called; the designated audio feature vector is further identified and processed based on the spam call recognition model to obtain the probability that the target phone belongs to each preset spam call category; finally, the target probability with the highest value is screened out from all the probabilities, and a spam call recognition result corresponding to the target phone is generated based on the target probability. When the target phone is connected to the present application, the initial audio data of the target phone is obtained, and the initial audio data is preprocessed to obtain the designated audio data, and then the designated audio data is feature extracted based on the use of the pre-trained language model to obtain the designated audio feature vector, and then the designated audio feature vector is identified and processed based on the use of the spam call recognition model, so that the efficient identification and accurate classification of spam calls can be automatically and accurately completed, thereby improving the recognition efficiency and recognition accuracy of spam calls. Compared with traditional audio fingerprint technology, this application has lower computational complexity and memory usage when processing large-scale data, and can adapt to the rapid changes in spam call patterns, thereby effectively improving the real-time and accuracy of spam call identification.

[0097] In some optional implementations, before step S204, the electronic device may further perform the following steps:

[0098] Get a pre-built, labeled spam call dataset.

[0099] In this embodiment, the specific implementation process of obtaining the pre-constructed labeled spam call data set will be further described in detail in subsequent specific embodiments of the present application, and will not be elaborated in detail here.

[0100] The spam call dataset is divided into a training set and a validation set.

[0101] In this embodiment, the spam call data set can be randomly divided into a training set and a validation set. The division ratio can be adjusted according to the specific task and the size of the data set. Usually, the training set accounts for the majority and the validation set accounts for a smaller portion. In addition, ensure that the data distribution of the training set and the validation set is consistent to avoid data imbalance. Specifically, data balance can be achieved by calculating the proportion of samples in each category and adjusting the division strategy appropriately.

[0102] Call a preset deep learning model and determine the target loss function and target optimization algorithm corresponding to the deep learning model.

[0103] In this embodiment, the above-mentioned deep learning model can select a deep learning network architecture suitable for audio classification tasks, such as convolutional neural network (CNN), recurrent neural network (RNN), etc. Among them, the cross entropy loss function can be selected as the target loss function of the deep learning model. The cross entropy loss function can well measure the difference between the model prediction result and the actual label, and is suitable for multi-classification tasks. During the model training process, the cross entropy loss value of the model on each batch of data is calculated. Then, according to the size of the loss value, the performance of the model can be evaluated and adjusted accordingly. In addition, according to the complexity of the model and the data scale, a suitable optimizer, such as Adam, SGD or RMSprop, can be selected. Preferably, the Adam optimizer usually has a good convergence speed and performance, and is suitable for most deep learning models. Among them, the relevant parameters of the optimizer, such as learning rate, momentum, etc., can be set according to actual needs. These parameters have a great influence on the training effect and convergence speed of the model, and can be adjusted according to specific tasks.

[0104] Based on the target loss function and the target optimization algorithm, the deep learning model is trained using the training set to obtain a corresponding first deep learning model.

[0105] In this embodiment, the process of training the deep learning model using the training set includes: initializing the weights and bias parameters of the deep learning model, which can be initialized randomly or by pre-training the model. Then the training set is divided into multiple batches, each batch containing a certain number of samples. On each batch, the output and loss value of the deep learning model are calculated, and the weights and bias parameters of the deep learning model are updated by the back propagation algorithm, thereby obtaining a trained first deep learning model.

[0106] The first deep learning model is adjusted based on the validation set to obtain a second deep learning model that meets preset performance requirements.

[0107] In this embodiment, after each training cycle (epoch), the performance of the first deep learning model is evaluated using the validation set. According to indicators such as the loss value and accuracy on the validation set, the generalization ability and performance of the first deep learning model are judged. Then, according to the evaluation results on the validation set, the hyperparameters of the first deep learning model, such as learning rate, batch size, number of network layers, etc., are adjusted. Through multiple attempts and adjustments, the optimal hyperparameter combination is found to improve the performance of the first deep learning model. When the performance of the first deep learning model on the validation set no longer improves or reaches the preset number of training rounds, the training process is stopped to obtain a second deep learning model that meets the performance requirements. In addition, the trained model weights and parameters can also be saved for subsequent model evaluation and deployment.

[0108] The second deep learning model is used as the spam call identification model.

[0109] In this embodiment, the trained spam call identification model can be integrated into the detection system of the telephone network to ensure that the spam call identification model can receive the audio feature vector and output the classification result. Among them, the interface between the spam call identification model and the system can be designed, including data input, model call and result output, etc., to ensure that data can be smoothly transmitted between systems. In addition, as the spam call strategy continues to change, the system will regularly update the model to adapt to the new spam call pattern. This includes collecting new spam call samples, retraining and fine-tuning the model, and updating the parameters of the clustering algorithm.

[0110] The present application obtains a pre-constructed and annotated spam call dataset, and divides the spam call dataset into a training set and a validation set; then calls a preset deep learning model, and determines the target loss function and target optimization algorithm corresponding to the deep learning model; then, based on the target loss function and the target optimization algorithm, uses the training set to train the deep learning model to obtain a corresponding first deep learning model; subsequently, based on the validation set, the first deep learning model is adjusted to obtain a second deep learning model that meets the preset performance requirements; finally, the second deep learning model is used as the spam call recognition model. The present application divides a pre-constructed and annotated spam call dataset into a training set and a validation set, and then calls a preset deep learning model; subsequently, based on the determined target loss function and target optimization algorithm, the training set and the validation set are used to train and adjust the deep learning model, respectively, so that a spam call recognition model that meets the performance requirements can be quickly and accurately constructed, thereby improving the construction efficiency of the spam call recognition model and ensuring the model effect of the generated spam call recognition model.

[0111] In some optional implementations of this embodiment, obtaining a pre-constructed labeled spam call dataset includes the following steps:

[0112] The pre-collected voice telephone data is pre-processed to obtain audio data, and the audio data is feature extracted to obtain an audio feature vector.

[0113] In this embodiment, a large amount of voice call data can be obtained from the telephone network, voice recording system or partners. Among them, it is necessary to ensure that the voice call data format is an audio file (such as WAV, MP3, etc.) or a real-time audio stream. In addition, data privacy regulations are observed during the collection process to ensure the legality of the data and user privacy. In addition, the process of pre-processing the voice call data to obtain audio data can refer to the above-mentioned specific processing process of pre-processing the initial audio data to obtain the corresponding specified audio data, which will not be described in detail here. And, the process of extracting features from the audio data to obtain an audio feature vector can refer to the above-mentioned specific processing process of extracting features from the specified audio data based on the pre-trained language model to obtain the corresponding specified audio feature vector, which will not be described in detail here.

[0114] The audio feature vector is clustered based on a preset target clustering algorithm to obtain a corresponding clustering result.

[0115] In this embodiment, the above-mentioned target clustering algorithm can specifically adopt a high-dimensional clustering algorithm, such as DBSCAN (density-based clustering algorithm). Algorithm parameters, such as neighborhood radius (Eps) and minimum number of samples (MinPts), can be set according to the distribution of audio feature vectors and clustering requirements. Specifically, the process of clustering audio feature vectors based on the target clustering algorithm includes: inputting the extracted audio feature vectors into the target clustering algorithm. The target clustering algorithm calculates the distance between audio feature vectors (such as Euclidean distance, cosine similarity, etc.), and divides the audio feature vectors into different clusters according to the distance metric. In addition, feature vectors that do not conform to the clustering pattern can also be identified by the target clustering algorithm and treated as outliers.

[0116] An abnormality analysis is performed on the clustering result to obtain a corresponding abnormality analysis result.

[0117] In this embodiment, by analyzing the clustering results, a cluster of calls playing the same or similar audio messages is identified from the audio data, and the identified outliers can be further analyzed to determine whether they are misdialed calls or real spam calls.

[0118] Based on the clustering result and the abnormal analysis result, misdialed call audio and spam call audio are screened out from the audio data.

[0119] In this embodiment, corresponding labels (such as spam calls, normal calls, etc.) can be assigned to the above audio data based on clustering results and outlier analysis, thereby obtaining corresponding misdialed call audio and spam call audio.

[0120] The spam call audio is transcribed to obtain the corresponding spam call text.

[0121] In this embodiment, the spam call audio contained in the above audio data is obtained, and then the spam call audio is transcribed using an automatic speech recognition (ASR) tool to output the spam call text in a readable string format. The transcribed spam call text can be further quality checked to ensure the accuracy of the transcription. For transcription results with low quality, manual correction or re-transcription can be considered.

[0122] The language recognition of the spam call text is performed based on a preset language recognition model to obtain a corresponding language recognition result.

[0123] In this embodiment, the above-mentioned language recognition model is a multi-language recognition model with language recognition function. By inputting the transcribed spam call text into the language recognition model, the language recognition model will recognize the language in the spam call text (language recognition result) and output it as a language code (such as en for English, zh for Chinese, etc.).

[0124] Based on the preset target language requirement and the language recognition result, the designated spam call audio in the non-target language is filtered out from the spam call audio to obtain the filtered target spam call audio.

[0125] In this embodiment, the target language requirement can be determined according to actual business needs. According to the target language requirement, the designated spam call audio in a non-target language can be filtered out from the spam call audio to obtain the final target spam call audio. Exemplarily, if the target language is English, all non-English audio in the spam call audio is filtered out. In addition, the spam call audio files identified after clustering and the corresponding transcribed texts can be classified and stored according to different categories (such as fraud, sales promotion, harassment, etc.). A separate directory can be created for each category, and the corresponding audio files and text files can be stored in the directory.

[0126] The target spam call audio is annotated based on a preset annotation strategy to obtain the spam call data set.

[0127] In this embodiment, a team of professional labelers can be formed to listen to the audio of each category of spam calls and label them. Specifically, audio labeling tools or customized labeling interfaces can be used for labeling. Among them, labeling specifications are formulated in advance to clarify the definition and labeling requirements of each label. For example, for fraudulent calls, labelers need to identify key information such as the type, means and target of the fraud. In addition, the labeling results can be quality checked to ensure the accuracy and consistency of the labeling. For samples with incorrect labeling, they need to be re-labeled or corrected.

[0128] The required spam call recognition model is trained and constructed by using the spam call dataset corresponding to the target language, so that the generated spam call recognition model can focus on the spam call analysis of a specific language (target language), thereby effectively improving the recognition effect of the spam call recognition model in the specific language.

[0129] The present application obtains audio data by preprocessing pre-collected voice call data, and extracts features from the audio data to obtain audio feature vectors; then clusters the audio feature vectors based on a preset target clustering algorithm to obtain corresponding clustering results; then performs an abnormality analysis on the clustering results to obtain corresponding abnormality analysis results; and based on the clustering results and the abnormality analysis results, screens out misdialed call audio and spam call audio from the audio data; subsequently transcribes the spam call audio to obtain corresponding spam call text; and further performs language recognition on the spam call text based on a preset language recognition model. The corresponding language recognition result is obtained; and based on the preset target language requirement and the language recognition result, the designated spam call audio in the non-target language is filtered out from the spam call audio to obtain the filtered target spam call audio; finally, the target spam call audio is annotated based on the preset annotation strategy to obtain the spam call data set. In this way, the present application performs data construction processing on the pre-collected voice call data based on the use of the target clustering algorithm, the language recognition model and the annotation strategy, which can realize the rapid and accurate construction of a spam call data set that meets the requirements, and effectively ensure the data accuracy and compliance of the obtained spam call data set. It is conducive to the subsequent training of the deep learning model based on the use of the spam call data set, and can effectively construct a spam call recognition model with good spam call recognition effect.

[0130] In some optional implementations, step S202 includes the following steps:

[0131] The audio data is subjected to denoising processing to obtain corresponding first audio data.

[0132] In this embodiment, a filter (such as a bandpass filter, a high-pass filter) or a noise suppression algorithm may be used to reduce the background noise of the audio data, and the filter parameters may be adjusted according to the audio quality to achieve the best denoising effect.

[0133] Perform voice activity detection processing on the first audio data to obtain corresponding second audio data.

[0134] In this embodiment, the VAD algorithm library can be used to determine the presence of voice activity by detecting the energy and spectrum characteristics of the audio signal, and separate the detected voice segments from the silent segments, retaining only the calls containing valid voice information. The sensitivity of VAD can be set according to actual needs to balance the false alarm rate and the missed alarm rate.

[0135] The second audio data is used as the designated audio data.

[0136] The present application performs denoising processing on the audio data to obtain corresponding first audio data; then performs voice activity detection processing on the first audio data to obtain corresponding second audio data; and subsequently uses the second audio data as the designated audio data. The present application performs denoising processing and voice activity detection processing on the initial audio data, thereby achieving fast and intelligent preprocessing of the initial audio data, and then accurately identifying and retaining the designated audio data containing valid voice information, thereby ensuring the data quality of subsequent processing and providing accurate input for feature extraction processing of the designated audio data.

[0137] In some optional implementations, step S203 includes the following steps:

[0138] The pre-trained speech model is called.

[0139] In this embodiment, the pre-trained speech model can specifically adopt a pre-trained language model based on the Transformer architecture, such as WavLM, which has been pre-trained on a large amount of human speech data and can capture the characteristics of human speech. The pre-trained speech model is loaded from the pre-trained model library to ensure that the model parameters are consistent with those during pre-training. In addition, the pre-trained speech model can be fine-tuned or adapted according to specific tasks (such as spam call recognition).

[0140] The specified audio data is forward-propagated based on the pre-trained language model to obtain a corresponding hidden layer state.

[0141] In this embodiment, after the specified audio data is input into the pre-trained language model, the pre-trained language model performs a forward propagation calculation on the input specified audio data to obtain a hidden layer state.

[0142] A mean pooling operation is performed on the hidden layer state to obtain a corresponding feature vector.

[0143] In this embodiment, a feature vector of a fixed dimension for each audio recording is obtained by performing a mean pooling operation on the obtained hidden layer state.

[0144] The specified audio feature vector is generated based on the feature vector.

[0145] In this embodiment, the obtained feature vector can be directly used as the above-mentioned designated audio feature vector. In addition, the extracted designated audio feature vector can also be stored in a database or used for subsequent processing.

[0146] The present application calls the pre-trained speech model; then performs forward propagation processing on the specified audio data based on the pre-trained language model to obtain the corresponding hidden layer state; then performs mean pooling operation on the hidden layer state to obtain the corresponding feature vector; and subsequently generates the specified audio feature vector based on the feature vector. The present application performs forward propagation processing on the specified audio data based on the use of a pre-trained language model to obtain the corresponding hidden layer state, and then performs mean pooling operation on the hidden layer state, so as to realize fast and accurate extraction of the required specified audio feature vector from the specified audio data, thereby ensuring the accuracy of the obtained specified audio feature vector, which is conducive to providing accurate feature input for the subsequent recognition processing of the spam call recognition model.

[0147] In some optional implementations of this embodiment, step S206 includes the following steps:

[0148] Get the preset probability threshold.

[0149] In this embodiment, a suitable probability threshold can be selected according to actual needs. This probability threshold is used to determine whether a call is a spam call. For example, a probability threshold can be set to 0.8, indicating that only when the probability of a certain spam call category exceeds 0.8, the call is considered to be a spam call of the spam call category. In addition, according to the performance evaluation results (such as accuracy, recall rate, etc.) of the spam call recognition model, the above probability threshold can be adjusted to optimize the classification effect.

[0150] Determine whether the target probability is greater than the probability threshold.

[0151] In this embodiment, the target probability with the highest probability value can be numerically compared with the above probability threshold to obtain a corresponding numerical comparison result, wherein the numerical comparison result includes: the target probability is greater than or equal to the probability threshold, or the target probability is less than the probability threshold.

[0152] If so, a spam call identification result is generated indicating that the target phone is a spam call.

[0153] In this embodiment, if the target probability is greater than or equal to the probability threshold, the target phone is determined to be a spam call, and a spam call identification result indicating that the target phone is a spam call is generated. In addition, the target spam call category (i.e., the spam call category corresponding to the highest probability) to which the target phone belongs can be further identified.

[0154] If not, a spam call identification result is generated indicating that the target phone is a normal phone.

[0155] In this embodiment, if the target probability is less than the probability threshold, the target phone is determined to be a normal phone, and a spam call identification result indicating that the target phone is a normal phone is generated.

[0156] The present application obtains a preset probability threshold; then determines whether the target probability is greater than the probability threshold; if so, generates a spam call identification result that the target phone is a spam call; if not, generates a spam call identification result that the target phone is a normal phone. The present application obtains a preset probability threshold, then compares the screened target probability with the probability threshold, and then based on the obtained numerical comparison result, can quickly and accurately generate a spam call identification result corresponding to the target phone, thereby improving the accuracy and intelligence of the generated spam call identification result.

[0157] In some optional implementations of this embodiment, before step S206, the electronic device may further perform the following steps:

[0158] If the content of the spam call identification result is that the target phone is a spam call, a target spam call category corresponding to the target probability is obtained.

[0159] In this embodiment, the target spam call category refers to the type of spam call to which the target phone belongs.

[0160] A call analysis result corresponding to the target call is generated based on the target spam call category and the spam call identification result.

[0161] In this embodiment, the target junk phone category and the junk phone identification result may be integrated to obtain integrated data, and the integrated data may be used as the phone analysis result of the target phone.

[0162] The telephone analysis result is outputted.

[0163] In this embodiment, the output processing may include recording the telephone analysis result into a database or outputting it into a designated file.

[0164] If the present application detects that the content of the spam call identification result is that the target phone is a spam call, the target spam call category corresponding to the target probability is obtained; then a phone analysis result corresponding to the target phone is generated based on the target spam call category and the spam call identification result; and the phone analysis result is subsequently output and processed. If the present application detects that the content of the spam call identification result is that the target phone is a spam call, the target spam call category corresponding to the target probability is intelligently obtained, and a phone analysis result corresponding to the target phone is automatically generated based on the target spam category and the spam call identification result, and then the phone analysis result is output and processed, thereby completing the data record processing of the phone analysis result, which is conducive to subsequent convenient data query.

[0165] In some optional implementations, the user information obtained is subject to the user's consent and complies with relevant laws and policies.

[0166] In addition, the non-Company software tools or components that appear in the embodiments of the present application are merely examples and do not represent actual use.

[0167] In addition, the spam call identification method based on audio feature embedding proposed in this application can effectively process large-scale telephone data and improve the accuracy and efficiency of spam call identification by utilizing advanced machine learning models and big data processing technologies. This application can not only quickly identify and classify spam calls, but also adapt to changes in spam call strategies and update recognition patterns in real time, thereby more effectively protecting users from spam calls. Moreover, this application can more comprehensively collect and analyze audio content when processing interactive spam calls, improve the ability to understand and identify spam call behavior, and can more comprehensively collect and analyze audio content in calls, improve the efficiency of extracting key information, and provide a more advanced and effective solution for the field of telecommunications security.

[0168] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0169] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned spam call identification results, the above-mentioned spam call identification results can also be stored in a node of a blockchain.

[0170] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0171] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0172] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0173] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0174] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0175] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a device for identifying spam calls. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0176] like Figure 3 As shown, the spam call identification device 300 described in this embodiment includes: a first acquisition module 301, a preprocessing module 302, an extraction module 303, a calling module 304, an identification module 305 and a first generation module 306. Among them:

[0177] The first acquisition module 301 is used to acquire initial audio data corresponding to the target phone when the target phone is accessed;

[0178] A preprocessing module 302 is used to preprocess the initial audio data to obtain corresponding specified audio data;

[0179] An extraction module 303 is used to extract features from the specified audio data based on a pre-trained language model to obtain a corresponding specified audio feature vector;

[0180] A calling module 304 is used to call a preset spam call identification model;

[0181] The identification module 305 is used to perform identification processing on the specified audio feature vector based on the spam call identification model to obtain the probability that the target phone belongs to each preset spam call category;

[0182] The first generating module 306 is used to select a target probability with the highest value from all the probabilities, and generate a spam call identification result corresponding to the target phone based on the target probability.

[0183] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the method for identifying spam calls in the above-mentioned implementation mode, and will not be described in detail here.

[0184] In some optional implementations of this embodiment, the apparatus for identifying spam calls further includes:

[0185] The second acquisition module is used to obtain a pre-built and labeled spam call dataset;

[0186] A partitioning module, used for partitioning the spam call data set into a training set and a validation set;

[0187] A first determination module is used to call a preset deep learning model and determine a target loss function and a target optimization algorithm corresponding to the deep learning model;

[0188] A training module, used to train the deep learning model using the training set based on the target loss function and the target optimization algorithm to obtain a corresponding first deep learning model;

[0189] An adjustment module, configured to adjust the first deep learning model based on the validation set to obtain a second deep learning model that meets preset performance requirements;

[0190] The second determination module is used to use the second deep learning model as the spam call identification model.

[0191] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the method for identifying spam calls in the above-mentioned implementation mode, and will not be described in detail here.

[0192] In some optional implementations of this embodiment, the second acquisition module includes:

[0193] A first processing submodule, configured to pre-process the pre-collected voice call data to obtain audio data, and extract features from the audio data to obtain an audio feature vector;

[0194] A clustering submodule, used to perform clustering processing on the audio feature vector based on a preset target clustering algorithm to obtain a corresponding clustering result;

[0195] An analysis submodule, used to perform an abnormality analysis on the clustering result to obtain a corresponding abnormality analysis result;

[0196] A screening submodule, for screening misdialed call audio and spam call audio from the audio data based on the clustering result and the abnormality analysis result;

[0197] A transcription submodule, used to transcribe the spam call audio to obtain the corresponding spam call text;

[0198] The recognition submodule is used to perform language recognition on the spam call text based on a preset language recognition model to obtain a corresponding language recognition result;

[0199] A filtering submodule, configured to filter out designated spam call audio in a non-target language from the spam call audio based on a preset target language requirement and the language recognition result, to obtain filtered target spam call audio;

[0200] The labeling submodule is used to label the target spam call audio based on a preset labeling strategy to obtain the spam call data set.

[0201] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the method for identifying spam calls in the above-mentioned implementation mode, and will not be described in detail here.

[0202] In some optional implementations of this embodiment, the preprocessing module 302 includes:

[0203] A second processing submodule, configured to perform denoising on the audio data to obtain corresponding first audio data;

[0204] A third processing submodule, configured to perform voice activity detection processing on the first audio data to obtain corresponding second audio data;

[0205] A determination submodule is used to use the second audio data as the designated audio data.

[0206] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the method for identifying spam calls in the above-mentioned implementation mode, and will not be described in detail here.

[0207] In some optional implementations of this embodiment, the extraction module 303 includes:

[0208] A calling submodule, used to call the pre-trained speech model;

[0209] A fourth processing submodule, configured to perform forward propagation processing on the specified audio data based on the pre-trained language model to obtain a corresponding hidden layer state;

[0210] A fifth processing submodule is used to perform a mean pooling operation on the hidden layer state to obtain a corresponding feature vector;

[0211] The first generating submodule is used to generate the specified audio feature vector based on the feature vector.

[0212] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the method for identifying spam calls in the above-mentioned implementation mode, and will not be described in detail here.

[0213] In some optional implementations of this embodiment, the first generating module 306 includes:

[0214] An acquisition submodule is used to obtain a preset probability threshold;

[0215] A judgment submodule, used to judge whether the target probability is greater than the probability threshold;

[0216] A second generating submodule is used to generate a spam call identification result indicating that the target phone is a spam call if yes;

[0217] The third generating submodule is used to generate a spam call identification result indicating that the target phone is a normal phone if no.

[0218] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the method for identifying spam calls in the above-mentioned implementation mode, and will not be described in detail here.

[0219] In some optional implementations of this embodiment, the apparatus for identifying spam calls further includes:

[0220] A third acquisition module, configured to acquire a target spam call category corresponding to the target probability if the content of the spam call identification result is that the target call is a spam call;

[0221] A second generating module, configured to generate a telephone analysis result corresponding to the target telephone based on the target spam call category and the spam call identification result;

[0222] The output module is used to output the telephone analysis result.

[0223] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the method for identifying spam calls in the above-mentioned implementation mode, and will not be described in detail here.

[0224] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0225] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components 41-43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0226] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0227] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. equipped on the computer device 4. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of the method for identifying spam calls, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0228] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run computer-readable instructions stored in the memory 41 or process data, such as computer-readable instructions for running the spam call identification method.

[0229] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0230] The present application also provides another embodiment, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the above-mentioned spam call identification method.

[0231] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0232] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A method for identifying spam calls, characterized in that: The steps include: When a target phone is accessed, initial audio data corresponding to the target phone is obtained; Preprocessing the initial audio data to obtain corresponding specified audio data; Perform feature extraction on the specified audio data based on the pre-trained language model to obtain a corresponding specified audio feature vector; Call the preset spam call identification model; Performing recognition processing on the designated audio feature vector based on the spam call recognition model to obtain the probability that the target phone belongs to each preset spam call category; A target probability with the highest value is screened out from all the probabilities, and a spam call identification result corresponding to the target phone is generated based on the target probability.

2. The method for identifying spam calls according to claim 1, characterized in that: Before the step of calling the preset spam call identification model, the method further includes: Get a pre-built, labeled spam call dataset; Dividing the spam call dataset into a training set and a validation set; Calling a preset deep learning model and determining a target loss function and a target optimization algorithm corresponding to the deep learning model; Based on the target loss function and the target optimization algorithm, the deep learning model is trained using the training set to obtain a corresponding first deep learning model; Based on the validation set, the first deep learning model is adjusted to obtain a second deep learning model that meets preset performance requirements; The second deep learning model is used as the spam call identification model.

3. The method for identifying spam calls according to claim 2, characterized in that: The step of obtaining a pre-constructed labeled spam call dataset specifically includes: Preprocessing the pre-collected voice call data to obtain audio data, and extracting features from the audio data to obtain an audio feature vector; Performing clustering processing on the audio feature vector based on a preset target clustering algorithm to obtain a corresponding clustering result; Performing an abnormality analysis on the clustering result to obtain a corresponding abnormality analysis result; Based on the clustering result and the abnormal analysis result, filtering out misdialed call audio and spam call audio from the audio data; Transcribing the spam call audio to obtain the corresponding spam call text; Performing language recognition on the spam call text based on a preset language recognition model to obtain a corresponding language recognition result; Based on the preset target language requirement and the language recognition result, filtering out the designated spam call audio in the non-target language from the spam call audio to obtain the filtered target spam call audio; The target spam call audio is annotated based on a preset annotation strategy to obtain the spam call data set.

4. The method for identifying spam calls according to claim 1, characterized in that: The step of preprocessing the initial audio data to obtain the corresponding specified audio data specifically includes: Performing denoising processing on the audio data to obtain corresponding first audio data; Performing voice activity detection processing on the first audio data to obtain corresponding second audio data; The second audio data is used as the designated audio data.

5. The method for identifying spam calls according to claim 1, characterized in that: The step of extracting features from the specified audio data based on the pre-trained language model to obtain a corresponding specified audio feature vector specifically includes: Calling the pre-trained speech model; Performing forward propagation processing on the specified audio data based on the pre-trained language model to obtain a corresponding hidden layer state; Performing a mean pooling operation on the hidden layer state to obtain a corresponding feature vector; The specified audio feature vector is generated based on the feature vector.

6. The method for identifying spam calls according to claim 1, characterized in that: The step of generating a spam call identification result corresponding to the target phone based on the target probability specifically includes: Obtaining a preset probability threshold; Determining whether the target probability is greater than the probability threshold; If yes, generating a spam call identification result indicating that the target phone is a spam call; If not, a spam call identification result is generated indicating that the target phone is a normal phone.

7. The method for identifying spam calls according to claim 1, characterized in that: After the step of generating a spam call identification result corresponding to the target phone based on the target probability, the method further includes: If the content of the spam call identification result is that the target phone is a spam call, obtaining a target spam call category corresponding to the target probability; Generate a call analysis result corresponding to the target call based on the target spam call category and the spam call identification result; The telephone analysis result is outputted.

8. A device for identifying spam calls, characterized in that: include: A first acquisition module, used for acquiring initial audio data corresponding to the target phone when the target phone is accessed; A preprocessing module, used for preprocessing the initial audio data to obtain corresponding specified audio data; An extraction module, used to extract features from the specified audio data based on a pre-trained language model to obtain a corresponding specified audio feature vector; A calling module, used to call a preset spam call identification model; An identification module, configured to perform identification processing on the specified audio feature vector based on the spam call identification model to obtain the probability that the target phone belongs to each preset spam call category; The first generating module is used to select a target probability with the highest value from all the probabilities, and generate a spam call identification result corresponding to the target phone based on the target probability.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the method for identifying spam calls according to any one of claims 1 to 7 when executing the computer-readable instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the method for identifying spam calls according to any one of claims 1 to 7 are implemented.

Images