FreeSwitch-based number state identification method and system, medium and electronic equipment

Abstract

The invention discloses a number state identification method based on FreeSwitch. The method comprises the following steps: constructing a data set; according to the constructed data set, a CNN framework is adopted, and a binary-classification early media detection model is obtained through training; using the early media detection model to identify the early media ringtone of the called user, and classifying the early media ringtone into a normal state or an abnormal state; the classification result is fed back to the FreeSwitch platform, and for the normal early media ringtone, the early media is pushed to the dialing user according to the original process; and for the abnormal early media ringtone, the corresponding abnormal state type is analyzed, and the communication line is released. According to the invention, based on the FreeSwitch platform, in the off-line network environment, the identification and classification of the number state in the call are realized by a simpler neural network structure, and the method has higher identification speed and higher identification accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of speech recognition, and in particular to a method for identifying number status based on FreeSwitch. Background Art

[0002] FreeSwitch is an open source real-time communication platform that supports various communication protocols and applications and has strong scalability. One of its important functions is automatic outbound calling, but there are many numbers that cannot be connected. Because early media will be played, the communication line will not be released immediately, and it will take a long time to automatically hang up, which will cause a great waste of line resources.

[0003] If early media can be effectively identified, the utilization rate of line resources will surely be improved. To achieve this goal, automatic speech recognition technology (ASR) has been used to identify early media. It converts the voice information of early media into text information, and then analyzes the semantics of the text information to determine the subsequent interaction processing method of the intelligent interactive system. However, since ASR technology is a technology that consumes a lot of machine resources, using this technology for early media identification will consume a lot of computing resources and increase business costs.

[0004] Convolutional Neural Network (CNN) is a deep learning model for processing structured data (especially images). CNN is widely used in computer vision tasks such as image classification, object detection, image segmentation, etc. The main feature of CNN is that it can automatically learn the features of input data while reducing computational complexity through shared weights and local connections. It is very suitable for classification prediction of early media. However, convolutional neural network is a classification model suitable for images. How to apply it to classification prediction of early media is an urgent problem to be solved. Summary of the invention

[0005] The present invention aims to solve one of the technical problems existing in the related art at least to a certain extent.

[0006] One object of the present invention is to provide a number status recognition method based on FreeSwitch, which uses a convolutional neural network to analyze early media sounds and obtain the current status of the calling number as soon as possible. For numbers in abnormal status, the line can be released in advance and the user can be informed to improve efficiency.

[0007] Another object of the present invention is to provide a FreeSwitch-based number status identification system.

[0008] In order to achieve the above-mentioned object, the present invention provides a method for identifying number status based on FreeSwitch, which is characterized by comprising:

[0009] Record the early media ringtones received by the FreeSwitch voice platform in segments, extract the Mel-frequency cepstral coefficients of the early media ringtones, and draw them into a Mel-frequency cepstral coefficient graph to construct a data set;

[0010] Based on the constructed dataset, a CNN framework is used to train a binary classification early media detection model;

[0011] The early media detection model is used to identify the early media ringtone of the called user and classify the early media ringtone as normal or abnormal;

[0012] The classification results are fed back to the FreeSwitch platform. The FreeSwitch platform is configured to push the early media to the dial-up user according to the original process for normal early media ringtones, waiting for the called user to connect; for abnormal early media ringtones, analyze the corresponding abnormal state type and release the communication line.

[0013] A further preferred technical solution of the present invention is that the early media ringtones received by the FreeSwitch voice platform are recorded in a fragmented manner, the Mel-cepstral coefficients of the early media ringtones are extracted, and a Mel-cepstral coefficient graph is plotted to construct a data set; including:

[0014] Collect early media ringtone files of the FreeSwitch voice platform, batch preprocess the early media ringtone files, and edit them into 2-second audio files to ensure that the edited audio files contain the first sound information in the early media ringtone, and give normal or abnormal labels based on the ringtone within 2 seconds after the audio has sound.

[0015] The Mel-cepstral coefficients are extracted from the edited audio file and the Mel-cepstral coefficient map is drawn. After scaling, the Mel-cepstral coefficient map is drawn into a 224×224 pixel RGB three-channel map. The map is matched with the label to construct the data set used to train CNN.

[0016] Preferably, the adopted CNN framework structure is 6 layers of convolution and pooling layers followed by 3 layers of fully connected layers.

[0017] Preferably, the step of identifying the early media ringtone of the called user by using the early media detection model further includes a step of detecting an empty tone, specifically:

[0018] On the FreeSwitch platform, the early media ringtone after the call is connected is recorded, and each 2-second segment is added to the queue for processing. Then, the 2-second audio segments in the queue are tested for empty sound one by one.

[0019] When more than 95% of the content in the audio clip is below -60dB, the audio clip is determined to be a blank sound clip, and then the blank sound detection is continued for the next recorded 2-second audio clip;

[0020] If it is not empty sound, the empty sound part before the sound in the audio clip is clipped, and the remaining sound part is returned and recognized using the early media detection model.

[0021] Preferably, the method of using an early media detection model to identify the early media ringtone of the called user and classifying the early media ringtone as normal or abnormal includes:

[0022] Extract the Mel cepstral coefficients from the processed non-blank audio clips and draw a Mel cepstral coefficient graph. After scaling the Mel cepstral coefficient graph, draw it into a 224×224 pixel RGB three-channel graph, and call the trained early media detection model for classification prediction.

[0023] Within 2 seconds after the early media ringtone sounds, if the early media ringtone is a ringing tone or a ringback tone waiting to be connected, it is defined as a normal state. If the early media prompt tone that cannot be connected appears within 2 seconds, it is defined as an abnormal state.

[0024] Preferably, the analyzing the corresponding abnormal state type for the abnormal early media ringtone includes:

[0025] Record the early media ringtone after dialing the phone on the FreeSwitch platform, with each 2 seconds being an audio clip;

[0026] After the audio clip is judged as abnormal by the early media detection model, a corresponding audio clip of 2 seconds is added;

[0027] Call faster-whisper's offline model to extract text from the appended audio clips, detect keywords for specific abnormal states, classify specific abnormal state types, and feed the results back to the FreeSwitch platform. FreeSwitch releases the communication line in advance and informs the dial-up user of the specific abnormal state type of the number they dialed.

[0028] As a preferred embodiment, the abnormal status types mainly include busy, unavailable, powered off, out of service, unable to connect and others;

[0029] Among them, non-mainstream abnormal status prompt tones and early media that cannot identify the five states of busy, empty number, powered off, out of service, and unable to connect are classified as other types.

[0030] Another aspect of the present invention provides a FreeSwitch-based number status identification system, comprising:

[0031] The data collection module is used to record the early media ringtones received by the FreeSwitch voice platform in segments, extract the Mel-frequency cepstral coefficients of the early media ringtones, and draw them into a Mel-frequency cepstral coefficient graph, based on which a data set is constructed;

[0032] The model building module is used to train a binary early media detection model using a CNN framework based on the constructed data set;

[0033] An anomaly detection module, used to identify the early media ringtone of the called user by using the early media detection model, and classify the early media ringtone as normal or abnormal;

[0034] The FreeSwitch access module is used to feed back the classification results to the FreeSwitch platform and set the FreeSwitch platform to push the early media to the dial-up user according to the original process for normal early media ringtones, waiting for the called user to connect; for abnormal early media ringtones, analyze the corresponding abnormal state type and release the communication line.

[0035] Another aspect of the present invention provides a non-transitory computer-readable storage medium having computer instructions stored thereon, the computer instructions enabling a computer to execute the above-mentioned FreeSwitch-based number status identification method.

[0036] On the other hand, the present invention provides an electronic device, comprising: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus, and the processor calls the logic instructions in the memory to execute the above-mentioned FreeSwitch-based number status identification method.

[0037] Beneficial effects: The present invention is based on the FreeSwitch platform. In an offline network environment, it realizes the recognition and classification of the number status in a call with a relatively simple neural network structure, and has a faster recognition speed and a higher recognition accuracy. The designed number status recognition module can achieve high concurrency operation at a lower cost and lower CPU resource occupancy, and can operate continuously and stably under high concurrency. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The figure is a flow chart of the FreeSwitch number status identification method according to the present invention.

[0039] Figure 2 It is a flowchart of CNN neural network training in the present invention.

[0040] Figure 3 It is a structural diagram of the CNN neural network in the present invention.

[0041] Figure 4 This is a graph of the Loss changes during iterative training of the CNN neural network in the present invention.

[0042] Figure 5 This is a test run time diagram of the number status identification module of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0044] Combine the following Figure 1-Figure 5 The present invention describes a FreeSwitch-based number status identification method, system, medium and electronic device.

[0045] In view of the problems raised in the background technology, in order to achieve the purpose of using convolutional neural network to recognize early media ringtones, Mel cepstral coefficients are extracted from the audio and drawn into an image, making it suitable as the input of the CNN neural network for classification prediction.

[0046] Accordingly, after a call is initiated on the FreeSwitch platform, CNN can be used to analyze early media sounds to obtain the current status of the number as soon as possible. For numbers in abnormal status, the line can be released in advance and the user can be informed to improve efficiency. The present invention is described in detail through the following embodiments.

[0047] Embodiment 1: This embodiment provides a method for identifying number status based on FreeSwitch, which is divided into two threads during processing.

[0048] In thread 1, the FreeSwitch platform records early media ringtone segments, saves them as a segment every two seconds, and adds the segment to a queue for processing.

[0049] Thread 2 uses the early media detection model to identify the early media ringtone of the called user with the audio clips in the queue. First, it judges that there is no sound. If more than 95% of the content in the clip is lower than -60dB, it is judged as no sound, and is fed back to FreeSwitch, waiting for the next audio clip. If the clip is not no sound, the content before the first sound in the audio clip is clipped, and the remaining content is transmitted to the early media detection model in the form of an audio array and the sampling rate. For example, in a 2-second audio clip, the first 1 second is no sound, and the last 1 second has sound, then it is judged that there is sound, and the empty sound part of the first second will be clipped, and the remaining 1 second of sound content will be passed to the early media detection model in the form of an audio array.

[0050] When the early media detection model performs detection, the mfcc (Mel Cepstral Coefficient) features are extracted from the received audio array, and a 224×224 pixel RGB image is drawn after scaling as the input of the early media detection model; the early media detection model passes the input RGB image through six layers of convolution and pooling, and then outputs the recognition result through three fully connected layers, and feeds the recognition result back to FreeSwitch.

[0051] In this embodiment, the early media detection model training process is as follows: Figure 2 As shown, the dataset used to train the CNN positive and abnormal binary classification model requires the collection of a large number of early media ringtones in advance and the preprocessing of the collected early media ringtones. The preprocessing method is as follows: the collected early media ringtone files are clipped into 2 to 3 seconds of audio files, and the clipped audio files are ensured to contain the information of the first sound in the early media ringtones, and the normal or abnormal labels are given according to the ringtones within 2 seconds after the audio has sound; the mfcc features are extracted from the clipped audio files and the spectrogram is drawn, and the spectrogram is scaled and drawn into a 224×224 pixel RGB three-channel image, and the image is matched with the label to construct the dataset used for training the neural network.

[0052] In this embodiment, the structure of the trained early media detection model is as follows: Figure 3 As shown in the figure, the convolution kernel size is 3×3, the stride is 1, and the padding is 1. After the convolution, the pooling is 2×2, the stride is 2, and the padding is 0. After 6 layers of convolution and pooling, the resulting graph is flattened to obtain a 3×3×512 vector, and the binary classification result is finally obtained through 3 layers of fully connected layers. The data set is divided into a training set and a validation set in a ratio of 8:2. After 10 cycles of training, the final accuracy on the validation set is 99.43%. The loss value decrease curve during the training process is shown in the figure below. Figure 4After loading the trained CNN model into the GPU, the model is called for classification prediction. Under the test condition of 100 concurrent tasks, each classification prediction task can give a classification result in about 2 seconds.

[0053] In this embodiment, if the recognition result is normal, the result is fed back to FreeSwitch, FreeSwitch will maintain the communication line, and the dial-up user will wait for the called user to be connected; if the recognition result is abnormal, the abnormal subdivision module will continue to be called, and the array of the next audio segment in the queue will be extracted and appended to the audio array processed in the previous step to form a new audio array containing specific abnormal status information, and the new audio array will be transmitted to the abnormal subdivision module. The abnormal subdivision module calls the base model of faster-whisper, extracts text from the audio, and then detects keywords corresponding to the abnormal status from the text to achieve the segmentation of the abnormal status. After the segmentation results are obtained, the results are fed back to the FreeSwitch platform, the module releases the memory, FreeSwitch releases the communication line in advance, and feeds back the abnormal status of the called user to the dial-up user.

[0054] Faster-whisper's existing audio-to-text extraction models include tiny, base, medium, and large. In actual tests, based on a comprehensive consideration of accuracy and text extraction speed, the base model was selected to implement the function of extracting text from audio.

[0055] In actual applications, the early media ringtones recorded by the FreeSwitch platform will have more or less noise, and the base model has a fast text extraction speed but limited accuracy. When using faster-whisper to extract text from the actual recorded audio, there will be homophonic characters, traditional Chinese characters, and typos. It is necessary to extract text in batches and filter high-frequency words to determine the keywords corresponding to the abnormal state and build an abnormal state dictionary. At the same time, in order to increase the accuracy, it is necessary to sort the abnormal states according to the frequency of occurrence to build an abnormal state dictionary. In addition, to increase the speed of searching for keywords, multiple keywords corresponding to the same abnormal state can be encapsulated in the form of a set.

[0056] In this embodiment, the FreeSwitch-based number status identification module will feed back the number status to FreeSwitch in the form of a tuple, for example, the silent early media will be fed back (0, "silent"), the normal will be fed back (1, "normal"), the prompt tone of being on a call and busy will be fed back (2, "on a call"), the prompt tone of being unable to connect will be fed back (3, "unable to connect"), the prompt tone of being powered off will be fed back (4, "powered off"), the prompt tone of an empty number or no registration will be fed back (5, "empty number"), the prompt tone of being shut down or suspended service will be fed back (6, "shut down"), and some prompt tones whose specific status cannot be determined and which are very rare will be fed back (7, "other").

[0057] Among them, the module has a recognition rate of 99.1% for normal number status and 98.2% for abnormal number status when performing positive and abnormal classification. When performing abnormal subdivision, the recognition rate of busy number is 97.5%, the recognition rate of unreachable number is 92.8%, the recognition rate of shutdown number is 99.3%, the recognition rate of empty number is 99.2%, and the recognition rate of shutdown number is 97.9%. The specific running test time is as follows: Figure 5 shown.

[0058] Embodiment 2: This embodiment provides a FreeSwitch-based number status identification system, including:

[0059] The data collection module is used to record the early media ringtones received by the FreeSwitch voice platform in segments, extract the Mel-frequency cepstral coefficients of the early media ringtones, and draw them into a Mel-frequency cepstral coefficient graph, based on which a data set is constructed;

[0060] The model building module is used to train a binary early media detection model using a CNN framework based on the constructed data set;

[0061] An anomaly detection module, used to identify the early media ringtone of the called user by using the early media detection model, and classify the early media ringtone as normal or abnormal;

[0062] The FreeSwitch access module is used to feed back the classification results to the FreeSwitch platform and call the setting module of the FreeSwitch platform to set the FreeSwitch platform to push the early media to the dial-up user according to the original process for normal early media ringtones and wait for the called user to connect; for abnormal early media ringtones, analyze the corresponding abnormal state type and release the communication line.

[0063] Embodiment 3: This embodiment provides a non-transitory computer-readable storage medium, on which computer instructions are stored, and the computer instructions enable a computer to execute a FreeSwitch number status identification method, the method comprising the following steps:

[0064] Record the early media ringtones received by the FreeSwitch voice platform in segments, extract the Mel-frequency cepstral coefficients of the early media ringtones, and draw them into a Mel-frequency cepstral coefficient graph to construct a data set;

[0065] Based on the constructed dataset, a CNN framework is used to train a binary classification early media detection model;

[0066] The early media detection model is used to identify the early media ringtone of the called user and classify the early media ringtone as normal or abnormal;

[0067] The classification results are fed back to the FreeSwitch platform. The FreeSwitch platform is configured to push the early media to the dial-up user according to the original process for normal early media ringtones, waiting for the called user to connect; for abnormal early media ringtones, analyze the corresponding abnormal state type and release the communication line.

[0068] Embodiment 4: This embodiment provides an electronic device, which may include: a processor, a communications interface, a memory, and a communication bus, wherein the processor, the communications interface, and the memory communicate with each other via the communication bus. The processor may call the logic instructions in the memory to perform FreeSwitch number status identification, and the method includes the following steps:

[0069] Record the early media ringtones received by the FreeSwitch voice platform in segments, extract the Mel-frequency cepstral coefficients of the early media ringtones, and draw them into a Mel-frequency cepstral coefficient graph to construct a data set;

[0070] Based on the constructed dataset, a CNN framework is used to train a binary classification early media detection model;

[0071] The early media detection model is used to identify the early media ringtone of the called user and classify the early media ringtone as normal or abnormal;

[0072] The classification results are fed back to the FreeSwitch platform. The FreeSwitch platform is configured to push the early media to the dial-up user according to the original process for normal early media ringtones, waiting for the called user to connect; for abnormal early media ringtones, analyze the corresponding abnormal state type and release the communication line.

[0073] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0074] Embodiment 5: This embodiment provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform FreeSwitch number status identification. The method includes the following steps:

[0075] Record the early media ringtones received by the FreeSwitch voice platform in segments, extract the Mel-frequency cepstral coefficients of the early media ringtones, and draw them into a Mel-frequency cepstral coefficient graph to construct a data set;

[0076] Based on the constructed dataset, a CNN framework is used to train a binary classification early media detection model;

[0077] The early media detection model is used to identify the early media ringtone of the called user and classify the early media ringtone as normal or abnormal;

[0078] The classification results are fed back to the FreeSwitch platform. The FreeSwitch platform is configured to push the early media to the dial-up user according to the original process for normal early media ringtones, waiting for the called user to connect; for abnormal early media ringtones, analyze the corresponding abnormal state type and release the communication line.

[0079] In summary, the present invention has the following advantages:

[0080] 1. Applicable to the number status identification scenario of the FreeSwitch platform.

[0081] 2. The complexity of deployment is greatly reduced. You only need to add a module based on FreeSwitch number status recognition and use the interface to interact with the FreeSwitch platform.

[0082] 3. Fast and accurate number status identification can be achieved at a low cost, and high concurrency and stable operation can be achieved.

[0083] 4. It can identify the number status, realize the early release of the communication line, and optimize the utilization of communication resources.

[0084] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0085] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying number status based on FreeSwitch, characterized in that: include: Record the early media ringtones received by the FreeSwitch voice platform in segments, extract the Mel-frequency cepstral coefficients of the early media ringtones, and draw them into a Mel-frequency cepstral coefficient graph to construct a data set; Based on the constructed dataset, a CNN framework is used to train a binary classification early media detection model; The early media detection model is used to identify the early media ringtone of the called user and classify the early media ringtone as normal or abnormal; The classification results are fed back to the FreeSwitch platform. The FreeSwitch platform is configured to push the early media to the dial-up user according to the original process for normal early media ringtones, waiting for the called user to connect; for abnormal early media ringtones, analyze the corresponding abnormal state type and release the communication line.

2. The FreeSwitch number status identification method according to claim 1, characterized in that: The method includes recording the early media ringtones received by the FreeSwitch voice platform in a fragmented manner, extracting the Mel-frequency cepstral coefficients of the early media ringtones, and drawing them into a Mel-frequency cepstral coefficient graph, thereby constructing a data set; including: Collect early media ringtone files of the FreeSwitch voice platform, batch preprocess the early media ringtone files, and edit them into 2-second audio files to ensure that the edited audio files contain the first sound information in the early media ringtone, and give normal or abnormal labels based on the ringtone within 2 seconds after the audio has sound. The Mel-cepstral coefficients are extracted from the edited audio file and the Mel-cepstral coefficient map is drawn. After scaling, the Mel-cepstral coefficient map is drawn into a 224×224 pixel RGB three-channel map. The map is matched with the label to construct the data set used to train CNN.

3. The FreeSwitch number status identification method according to claim 2, characterized in that: The CNN framework structure adopted is 6 layers of convolution and pooling layers followed by 3 layers of fully connected layers.

4. The FreeSwitch number status identification method according to claim 1, characterized in that: The step of using the early media detection model to identify the early media ringtone of the called user also includes a step of detecting an empty tone, specifically: On the FreeSwitch platform, the early media ringtone after the call is connected is recorded, and each 2-second segment is added to the queue for processing. Then, the 2-second audio segments in the queue are tested for empty sound one by one. When more than 95% of the content in the audio clip is below -60dB, the audio clip is determined to be a blank sound clip, and then the blank sound detection is continued for the next recorded 2-second audio clip; If it is not empty sound, the empty sound part before the sound in the audio clip is clipped, and the remaining sound part is returned and recognized using the early media detection model.

5. The FreeSwitch number status identification method according to claim 4, characterized in that: The method of using the early media detection model to identify the early media ringtone of the called user and classifying the early media ringtone as normal or abnormal includes: Extract the Mel cepstral coefficients from the processed non-blank audio clips and draw a Mel cepstral coefficient graph. After scaling the Mel cepstral coefficient graph, draw it into a 224×224 pixel RGB three-channel graph, and call the trained early media detection model for classification prediction. Within 2 seconds after the early media ringtone sounds, if the early media ringtone is a ringing tone or a ringback tone waiting to be connected, it is defined as a normal state. If the early media prompt tone that cannot be connected appears within 2 seconds, it is defined as an abnormal state.

6. The FreeSwitch number status identification method according to claim 1, characterized in that: The analyzing of the abnormal state type of the abnormal early media ringtone includes: Record the early media ringtone after dialing the phone on the FreeSwitch platform, with each 2 seconds being an audio clip; After the audio clip is judged as abnormal by the early media detection model, a corresponding audio clip of 2 seconds is added; Call faster-whisper's offline model to extract text from the appended audio clips, detect keywords for specific abnormal states, classify specific abnormal state types, and feed the results back to the FreeSwitch platform. FreeSwitch releases the communication line in advance and informs the dial-up user of the specific abnormal state type of the number they dialed.

7. The FreeSwitch number status identification method according to claim 6, characterized in that: Abnormal status types mainly include busy, unavailable, powered off, out of service, unable to connect and others; Among them, non-mainstream abnormal status prompt tones and early media that cannot identify the five states of busy, empty number, powered off, out of service, and unable to connect are classified as other types.

8. A FreeSwitch-based number status identification system, characterized in that: include: The data collection module is used to record the early media ringtones received by the FreeSwitch voice platform in segments, extract the Mel-frequency cepstral coefficients of the early media ringtones, and draw them into a Mel-frequency cepstral coefficient graph, based on which a data set is constructed; The model building module is used to train a binary early media detection model using a CNN framework based on the constructed data set; An anomaly detection module, used to identify the early media ringtone of the called user by using the early media detection model, and classify the early media ringtone as normal or abnormal; The FreeSwitch access module is used to feed back the classification results to the FreeSwitch platform and call the setting module of the FreeSwitch platform to set the FreeSwitch platform to push the early media to the dial-up user according to the original process for normal early media ringtones and wait for the called user to connect; for abnormal early media ringtones, analyze the corresponding abnormal state type and release the communication line.

9. A non-transitory computer-readable storage medium having computer instructions stored thereon, wherein the computer instructions enable a computer to execute the FreeSwitch number status identification method according to any one of claims 1 to 7.

10. An electronic device comprising: A processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus, and the processor calls the logic instructions in the memory to execute the FreeSwitch number status identification method according to any one of claims 1 to 7.

Images