Voice single-pass detection method and device, electronic equipment, storage medium and computer program product

Abstract

The application provides a voice single-pass detection method, device, equipment, medium and product, and belongs to the technical field of artificial intelligence. The method comprises the following steps: acquiring voice session data packets; splitting the voice session data packets according to data flow directions to obtain multiple data flows; determining the missing conditions of real-time transport protocol (RTP) data packets and the missing conditions of real-time transport control protocol (RTCP) data packets of each data flow; determining the data type of the voice session data packets based on the missing conditions of the RTP data packets and the missing conditions of the RTCP data packets of each data flow; and performing single-pass detection on the voice session data packets based on the data type to obtain a single-pass detection result. The voice single-pass detection method provided by the application can reduce the influence of the missing of part of field data on the detection of the entire voice session data packet, guarantee the voice single-pass detection coverage, and thus improve the accuracy of voice single-pass detection.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, medium, and product for one-way voice detection. Background Technology

[0002] With the rapid growth of voice communication transmission over LTE networks, one-way communication issues during voice calls are becoming increasingly prominent. One-way voice communication typically refers to a situation where sound can be transmitted and received in only one direction during a voice call, while sound cannot be transmitted in the other direction. Therefore, one-way voice communication has a significant impact on user experience.

[0003] In existing technologies, one-way communication detection is performed based on the frame loss ratio of the Real-time Transport Protocol (RTP) / Real-time Transport Control Protocol (RTCP) packet header detection information. This method is commonly used for everyday user voice one-way recognition because it does not require RTP / RTCP voice payload parsing. However, this method requires diagnosis based on various measurement data of the RTP / RTCP packet header, which has high requirements for data quality. Currently, the RTP / RTCP measurement fields in the session data collected by Deep Packet Inspection (DPI) on the live network are often missing. These session data with missing RTP / RTCP measurement fields cannot be used for one-way recognition based on existing technologies, which limits the coverage of this method and results in low accuracy of voice one-way detection. Summary of the Invention

[0004] This invention provides a method, apparatus, device, medium, and product for voice one-way detection, in order to solve the problem of low accuracy in existing voice one-way detection technologies.

[0005] In a first aspect, the present invention provides a method for detecting one-way speech, comprising:

[0006] Obtain voice session data packets;

[0007] The voice session data packets are split according to the data flow direction to obtain multiple data streams;

[0008] Determine the missing Real-Time Transport Protocol (RTP) packets and Real-Time Transport Control Protocol (RTCP) packets for each data stream.

[0009] Based on the missing RTP packets and missing RTCP packets in each data stream, the data type of the voice session data packets is determined.

[0010] Based on the data type, one-way detection is performed on the voice session data packets to obtain the one-way detection result.

[0011] In one embodiment, determining the data type of the voice session data packets based on the missing RTP data packets and RTCP data packets of each data stream includes:

[0012] If there are no missing RTP packets in all data streams and no missing RTCP packets in all data streams, then the data type of the voice session data packet is determined to be the first type.

[0013] If all RTP packets in all data streams are complete and at least one RTCP packet in a data stream is missing, then the data type of the voice session data packet is determined to be the second type.

[0014] If at least one data stream has missing RTP packets, and all data streams have no missing RTCP packets, then the data type of the voice session data packets is determined to be the third type.

[0015] If at least one segment of the RTP data packets is missing, and at least one segment of the RTCP data packets is missing, then the data type of the voice session data packets is determined to be the fourth type.

[0016] In one embodiment, the step of performing one-way detection on the voice session data packets based on the data type to obtain the one-way detection result includes:

[0017] If the data type is the first type, the voice conversation data packet is input into the first single-pass detection model to obtain the first single-pass detection result output by the first single-pass detection model; the first single-pass detection model is obtained by training the model based on the first type of conversation data packet sample and the conversation single-pass label of the first type of conversation data packet sample.

[0018] If the data type is the second type, the voice conversation data packet is input into the second single-pass detection model to obtain the second single-pass detection result output by the second single-pass detection model; the second single-pass detection model is obtained by training the model based on the second type of conversation data packet sample and the conversation single-pass label of the second type of conversation data packet sample.

[0019] If the data type is the third type, then the multi-segment data stream is input into the third single-pass detection model to obtain the third single-pass detection result output by the third single-pass detection model; the third single-pass detection model is obtained by training the model based on the multi-segment data stream samples of the third type session data packet samples and the session single-pass labels of the third type session data packet samples;

[0020] If the data type is the fourth type, then the multi-segment data stream is input into the fourth single-pass detection model to obtain the fourth single-pass detection result output by the fourth single-pass detection model; the fourth single-pass detection model is obtained by training the model based on the multi-segment data stream samples of the fourth type session data packet samples and the session single-pass labels of the fourth type session data packet samples.

[0021] In one embodiment, the multiple data streams include a calling side uplink data stream, a calling side downlink data stream, a called side uplink data stream, and a called side downlink data stream.

[0022] In one embodiment, the first single-pass detection model, the second single-pass detection model, the third single-pass detection model, and the fourth single-pass detection model are obtained by training in the following manner:

[0023] Obtain voice session data packet samples;

[0024] The voice session data packet sample is split according to the data flow direction to obtain multiple data flow samples;

[0025] Determine the call type label for each data stream sample; the call type label includes a one-way label and a non-one-way label;

[0026] Based on the call type label of each data stream sample, determine the session one-way label of the voice session data packet sample;

[0027] Based on the missing RTP packets and missing RTCP packets in each data stream sample, the data type of the voice session data packet sample is determined.

[0028] Based on the voice session data packet samples, the data type, the session one-way label, and the multi-segment data stream samples, the boosting tree model is trained to obtain the first one-way detection model, the second one-way detection model, the third one-way detection model, and the fourth one-way detection model, respectively.

[0029] In one embodiment, the multiple data stream samples include calling-side uplink data stream samples, calling-side downlink data stream samples, called-side uplink data stream samples, and called-side downlink data stream samples; determining the session-only label of the voice session data packet sample based on the call type label of each data stream sample includes:

[0030] If the call type label of the calling side uplink data stream sample, the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the first session single-pass label.

[0031] If the call type label of the calling side uplink data stream sample and the called side downlink data stream sample are both the single-pass label, and the call type label of the calling side downlink data stream sample and the called side uplink data stream sample are both the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the second session single-pass label.

[0032] If the call type label of the calling side downlink data stream sample and the called side uplink data stream sample are both the single-pass label, and the call type label of the calling side uplink data stream sample and the called side downlink data stream sample are both the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the third session single-pass label.

[0033] If the call type label of the calling side downlink data stream sample is the single-pass label, and the call type labels of the calling side uplink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the non-single-pass labels, then the session single-pass label of the voice session data packet sample is determined to be the fourth session single-pass label.

[0034] If the call type label of the called side downlink data stream sample is the single-pass label, and the call type labels of the calling side uplink data stream sample, the calling side downlink data stream sample, and the called side uplink data stream sample are all the non-single-pass labels, then the session single-pass label of the voice session data packet sample is determined to be the fifth session single-pass label.

[0035] If the call type labels of the calling side downlink data stream sample and the called side downlink data stream sample are both the single-pass label, and the call type labels of the calling side uplink data stream sample and the called side uplink data stream sample are both the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the sixth session single-pass label.

[0036] If the call type label of the calling side uplink data stream sample, the calling side downlink data stream sample, and the called side downlink data stream sample are all the single-pass label, and the call type label of the called side uplink data stream sample is the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the seventh session single-pass label.

[0037] If the call type label of the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the single-pass label, and the call type label of the calling side uplink data stream sample is the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the eighth session single-pass label.

[0038] If the call type label of the calling side uplink data stream sample, the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the ninth session single-pass label.

[0039] In one embodiment, training the boosting tree model based on the voice session data packet samples, the data type, the session one-way label, and the multi-segment data stream samples to obtain a first one-way detection model, a second one-way detection model, a third one-way detection model, and a fourth one-way detection model includes:

[0040] The voice conversation data packet sample belonging to the first type is identified as the first type of conversation data packet sample; based on the first type of conversation data packet sample and the conversation one-way label of the first type of conversation data packet sample, the boosting tree model is trained to obtain the first one-way detection model;

[0041] The voice conversation data packet samples belonging to the second type are identified as second type conversation data packet samples; based on the second type conversation data packet samples and the conversation one-way labels of the second type conversation data packet samples, the boosting tree model is trained to obtain the second one-way detection model;

[0042] The voice conversation data packet sample belonging to the third type is identified as the third type conversation data packet sample; the data stream sample in the multi-segment data stream sample of the third type conversation data packet sample that does not have missing RTP data packets is identified as the first target data stream sample; the boosting tree model is trained based on the conversation one-way label of the first target data stream sample and the third type conversation data packet sample to obtain the third one-way detection model;

[0043] The voice conversation data packet sample belonging to the fourth type is identified as the fourth type conversation data packet sample; the data stream sample in the multi-segment data stream sample of the fourth type conversation data packet sample that does not have missing RTP data packets is identified as the second target data stream sample; based on the conversation one-way label of the second target data stream sample and the fourth type conversation data packet sample, the lifting tree model is trained to obtain the fourth one-way detection model.

[0044] Secondly, the present invention also provides a voice one-way detection device, comprising:

[0045] The acquisition module is used to acquire voice session data packets;

[0046] The splitting module is used to split the voice session data packets according to the data flow direction to obtain multiple data streams;

[0047] The first determining module is used to determine the missing status of Real-Time Transport Protocol (RTP) data packets and Real-Time Transport Control Protocol (RTCP) data packets in each data stream.

[0048] The second determining module is used to determine the data type of the voice session data packet based on the missing RTP data packets and the missing RTCP data packets of each data stream.

[0049] The single-pass detection module is used to perform single-pass detection on the voice session data packets based on the data type, and obtain the single-pass detection result.

[0050] Thirdly, the present invention provides an apparatus comprising an electronic device, the electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the above-described voice one-way detection methods.

[0051] Fourthly, the present invention also provides a medium comprising a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described voice one-way detection methods.

[0052] Fifthly, the present invention also provides a product comprising a computer program product, the computer program product comprising a computer program, the computer program being able to be stored on a non-transitory computer-readable storage medium, the computer program being executed by the processor to implement the steps of any of the above-described voice one-way detection methods.

[0053] The voice one-way detection method, apparatus, device, medium, and product provided by this invention splits the acquired voice session data packets according to the data flow direction to obtain multiple data streams. These streams are then fused with multi-dimensional features for voice one-way detection. Furthermore, based on the missing data packets of the Real-Time Transmission Protocol (RTP) and Real-Time Transmission Control Protocol (RTC) in each data stream, the data type of the voice session data packets is determined, enabling the classification of voice session data packets with different missing data conditions. One-way detection is then performed on these data packets based on their different missing data conditions to obtain the one-way detection results. This reduces the impact of missing data in some fields on the detection of the entire voice session data packet, ensuring the coverage of voice one-way detection and thus improving the accuracy of voice one-way detection. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0055] Figure 1 This is a flowchart illustrating the one-way voice detection method provided by the present invention;

[0056] Figure 2 This is a schematic diagram of the structure of the voice one-way detection device provided by the present invention;

[0057] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0059] The terms "first," "second," etc., used in this invention are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein.

[0060] The following is combined Figures 1-3 This invention describes the voice one-way detection method, apparatus, equipment, medium, and product provided by the present invention.

[0061] Figure 1 This is a flowchart illustrating the one-way voice detection method provided by the present invention.

[0062] like Figure 1 As shown, the voice one-way detection method provided by the present invention includes, but is not limited to, the following steps:

[0063] Step 100: Obtain the voice session data packet;

[0064] Step 200: Split the voice session data packet according to the data flow direction to obtain multiple data streams;

[0065] Step 300: Determine the missing Real-Time Transport Protocol (RTP) packets and the missing Real-Time Transport Control Protocol (RTCP) packets for each data stream.

[0066] Step 400: Determine the data type of the voice session data packet based on the missing RTP data packets and the missing RTCP data packets in each data stream;

[0067] Step 500: Based on the data type, perform one-way detection on the voice session data packet to obtain the one-way detection result.

[0068] It should be noted that the voice one-way detection method provided in this embodiment of the invention is implemented based on a voice one-way detection device. The method first splits the acquired voice session data packets into multiple data streams according to the data flow direction. Based on multi-dimensional features of the data streams, it performs RTP / RTCP packet integrity rate analysis to classify the data types of the voice session data packets. Then, according to different data types, it performs one-way detection on the voice session data packets using corresponding and adapted one-way detection models to obtain the one-way detection results. Compared with traditional one-way detection methods based on rules such as packet loss rate, the detection accuracy and efficiency are improved. The voice session data packet classification and multi-model selection method ensure that voice session data packets with low RTP / RTCP packet integrity rates can also be subject to one-way detection, improving the coverage of one-way detection. Furthermore, it has strong generalization capabilities and can be applied to various DPI data value mining and analysis. Therefore, this embodiment of the invention uses a voice one-way detection device as the execution subject to describe the voice one-way detection method.

[0069] Specifically, the voice one-way detection device can collect voice session data packets by calling the operator's DPI system. The voice session data packets include the collected voice session data measurement values, and more specifically, the uplink data flow measurement values ​​and downlink data flow measurement values ​​collected by the calling and called side collection points.

[0070] It should be noted that the voice session data measurement values, uplink data flow measurement values, and downlink data flow measurement values ​​all contain a large number of measurement values. In practical applications, key measurement values ​​can be extracted from all measurement values ​​based on business experience. Furthermore, the missing RTP / RTCP data packets can be determined based on these key measurement values. In one embodiment, as shown in Table 1, Table 1 is a key measurement value table containing 10 key measurement values, including key measurement values ​​such as RTP / RTCP data packet loss, latency, and jitter, which can be used to determine the missing RTP / RTCP data packets.

[0071] Table 1 Key Measurement Values

[0072]

[0073] Furthermore, the voice one-way detection device splits the voice session data packets according to the data flow direction to obtain multiple data streams. These multiple data streams include the calling side uplink data stream, the calling side downlink data stream, the called side uplink data stream, and the called side downlink data stream. The calling side uplink data stream refers to the data stream sent from the calling device to the network, the calling side downlink data stream refers to the data stream sent from the network to the calling device, the called side uplink data stream refers to the data stream sent from the called device to the network, and the called side downlink data stream refers to the data stream sent from the network to the called device. In addition, each data stream includes key measurement values ​​such as RTP / RTCP packet loss, latency, and jitter.

[0074] Furthermore, the voice one-way detection device determines the missing status of RTP data packets in each data stream based on key measurement values ​​such as packet loss, latency, and jitter of RTP data packets in each data stream, and determines the missing status of RTCP data packets in each data stream based on key measurement values ​​such as packet loss, latency, and jitter of RTCP data packets in each data stream.

[0075] Therefore, it can be understood that the voice single-pass detection device determines the missing RTP data packets of the calling side uplink data stream, the calling side downlink data stream, the called side uplink data stream, and the called side downlink data stream, and also determines the missing RTCP data packets of the calling side uplink data stream, the calling side downlink data stream, the called side uplink data stream, and the called side downlink data stream, respectively.

[0076] Furthermore, the voice one-way detection device determines the data type of the voice session data packets based on the missing RTP data packets and the missing RTCP data packets in each data stream.

[0077] Furthermore, the voice one-way detection device performs one-way detection on the voice session data packets based on the data type of the voice session data packets, and obtains the one-way detection result of the voice session data packets.

[0078] It should be noted that the results of one-way communication detection of voice session data packets can help network optimization personnel accurately detect voice services experiencing one-way communication, effectively assisting them in identifying one-way communication scenarios in the current network and reducing user complaints. After confirming the one-way communication detection results, network optimization personnel can diagnose and troubleshoot communication network issues, as well as optimize the communication network to improve user experience and satisfaction.

[0079] The voice one-way detection method provided by this invention splits the acquired voice session data packets according to the data flow direction to obtain multiple data streams. Multi-dimensional features are then fused for voice one-way detection. Furthermore, based on the missing data packets of the Real-Time Transmission Protocol (RTP) and Real-Time Transmission Control Protocol (RTC) in each data stream, the data type of the voice session data packets is determined, achieving classification of voice session data packets with different missing data conditions. One-way detection is then performed on voice session data packets with different missing data conditions to obtain the one-way detection results. This method can reduce the impact of missing data in some fields on the detection of the entire voice session data packet, ensuring the coverage of voice one-way detection and thus improving the accuracy of voice one-way detection.

[0080] Further, based on step 400, determining the data type of the voice session data packet based on the missing RTP data packets and RTCP data packets of each data stream includes:

[0081] If there are no missing RTP packets in all data streams and no missing RTCP packets in all data streams, then the data type of the voice session data packet is determined to be the first type.

[0082] If all RTP packets in all data streams are complete and at least one RTCP packet in a data stream is missing, then the data type of the voice session data packet is determined to be the second type.

[0083] If at least one data stream has missing RTP packets, and all data streams have no missing RTCP packets, then the data type of the voice session data packets is determined to be the third type.

[0084] If at least one segment of the RTP data packets is missing, and at least one segment of the RTCP data packets is missing, then the data type of the voice session data packets is determined to be the fourth type.

[0085] Specifically, if there are no missing RTP packets in all data streams and no missing RTCP packets in all data streams, the voice one-way detection device determines that the data type of the voice session data packet is the first type.

[0086] Furthermore, if all RTP packets in all data streams are complete, but at least one RTCP packet in a data stream is missing, the voice one-way detection device determines that the data type of the voice session packet is the second type.

[0087] Furthermore, if at least one data stream has missing RTP packets, but all data streams have no missing RTCP packets, the voice one-way detection device determines that the data type of the voice session packets is the third type.

[0088] Furthermore, if at least one segment of the data stream has missing RTP packets and at least one segment of the data stream has missing RTCP packets, the voice one-way detection device determines that the data type of the voice session packets is the fourth type.

[0089] Furthermore, if all RTP packets for the data streams are missing, the voice one-way detection device determines that the data type of the voice session data packet is type 5. The device then discards the voice session data packets belonging to type 5. It should be noted that if all RTP packets for a voice session data packet belonging to type 5 are missing, it indicates that the voice session data packet is abnormal and cannot be detected for one-way communication; therefore, it needs to be discarded.

[0090] This invention determines the data type of voice session data packets based on the missing RTP and RTCP data packets of each data stream, classifying voice session data packets with different missing data conditions. Then, it performs one-way detection based on these missing data packets to obtain the one-way detection results. This reduces the impact of missing data in some fields on the detection of the entire voice session data packet, ensuring the coverage of one-way voice detection and thus improving the accuracy of one-way voice detection.

[0091] Further, based on step 500, the step of performing one-way detection on the voice session data packets based on the data type to obtain a one-way detection result includes:

[0092] If the data type is the first type, the voice conversation data packet is input into the first single-pass detection model to obtain the first single-pass detection result output by the first single-pass detection model; the first single-pass detection model is obtained by training the model based on the first type of conversation data packet sample and the conversation single-pass label of the first type of conversation data packet sample.

[0093] If the data type is the second type, the voice conversation data packet is input into the second single-pass detection model to obtain the second single-pass detection result output by the second single-pass detection model; the second single-pass detection model is obtained by training the model based on the second type of conversation data packet sample and the conversation single-pass label of the second type of conversation data packet sample.

[0094] If the data type is the third type, then the multi-segment data stream is input into the third single-pass detection model to obtain the third single-pass detection result output by the third single-pass detection model; the third single-pass detection model is obtained by training the model based on the multi-segment data stream samples of the third type session data packet samples and the session single-pass labels of the third type session data packet samples;

[0095] If the data type is the fourth type, then the multi-segment data stream is input into the fourth single-pass detection model to obtain the fourth single-pass detection result output by the fourth single-pass detection model; the fourth single-pass detection model is obtained by training the model based on the multi-segment data stream samples of the fourth type session data packet samples and the session single-pass labels of the fourth type session data packet samples.

[0096] Specifically, if the data type of the voice session data packet is the first type, the voice one-way detection device inputs the voice session data packet into the first one-way detection model to obtain the first one-way detection result output by the first one-way detection model. The first one-way detection model is obtained by training the model based on the first type of session data packet sample and the session one-way label of the first type of session data packet sample.

[0097] Furthermore, if the data type of the voice session data packet is the second type, the voice one-way detection device inputs the voice session data packet into the second one-way detection model to obtain the second one-way detection result output by the second one-way detection model. The second one-way detection model is obtained by training the model based on the second type session data packet sample and the session one-way label of the second type session data packet sample.

[0098] Furthermore, if the data type of the voice session data packet is the third type, the voice one-way detection device inputs the voice session data packet into the third one-way detection model to obtain the third one-way detection result output by the third one-way detection model. The third one-way detection model is obtained by training the model based on the multi-segment data stream samples of the third type session data packet samples and the session one-way labels of the third type session data packet samples.

[0099] Furthermore, if the data type of the voice session data packet is the fourth type, the voice one-way detection device inputs the voice session data packet into the fourth one-way detection model to obtain the fourth one-way detection result output by the fourth one-way detection model. The fourth one-way detection model is obtained by training the model based on the multi-segment data stream samples of the fourth type session data packet samples and the session one-way labels of the fourth type session data packet samples.

[0100] According to the embodiments of the present invention, the corresponding single-pass detection model is called to perform single-pass detection based on the voice conversation data packets with different missing conditions, and the corresponding single-pass detection results are obtained. This can reduce the impact of missing data in some fields on the detection of the entire voice conversation data packet, ensure the coverage of voice single-pass detection, and thus improve the accuracy of voice single-pass detection.

[0101] Furthermore, the first single-pass detection model, the second single-pass detection model, the third single-pass detection model, and the fourth single-pass detection model are obtained through training in the following manner:

[0102] Obtain voice session data packet samples;

[0103] The voice session data packet sample is split according to the data flow direction to obtain multiple data flow samples;

[0104] Determine the call type label for each data stream sample; the call type label includes a one-way label and a non-one-way label;

[0105] Based on the call type label of each data stream sample, determine the session one-way label of the voice session data packet sample;

[0106] Based on the missing RTP packets and missing RTCP packets in each data stream sample, the data type of the voice session data packet sample is determined.

[0107] Based on the voice session data packet samples, the data type, the session one-way label, and the multi-segment data stream samples, the boosting tree model is trained to obtain the first one-way detection model, the second one-way detection model, the third one-way detection model, and the fourth one-way detection model, respectively.

[0108] Specifically, the voice one-way detection device acquires voice session data packet samples. It should be noted that, in order to obtain accurate voice session data packet samples, 5G new radio voice service dialing tests are conducted in a specific wireless network environment to capture the voice communication traffic passing through the network. The captured voice communication traffic is then analyzed and identified in depth by the DPI system to obtain training samples, which are the voice session data packet samples.

[0109] Furthermore, the voice one-way detection device splits the voice session data packet sample according to the data flow direction to obtain multiple data flow samples, which include the calling side uplink data flow sample, the calling side downlink data flow sample, the called side uplink data flow sample, and the called side downlink data flow sample.

[0110] Furthermore, the voice one-way detection device determines the call type label for each data stream sample, wherein the call type label includes a one-way label and a non-one-way label.

[0111] Furthermore, the voice one-way detection device determines the session one-way label of the voice session data packet sample based on the call type label of each data stream sample.

[0112] Furthermore, the voice one-way detection device determines the data type of the voice session data packet sample based on the missing RTP data packets and the missing RTCP data packets in each data stream sample.

[0113] Furthermore, the voice one-way detection device trains the eXtreme Gradient Boosting (XGBoost) model based on voice session data packet samples, the data type of voice session data packet samples, the session one-way labels of voice session data packet samples, and the multi-segment data stream samples of voice session data packet samples, to obtain a one-way detection model.

[0114] This invention, through the splitting, labeling, and classification of voice session data packet samples, determines the session one-way label and data type of the voice session data packet samples, and then trains a boosting tree model to achieve voice one-way detection, thereby improving the model's generalization ability. The trained one-way detection model can achieve one-way detection of different types of voice call data packets, realizing large-scale and intelligent detection of voice service one-way, and providing a reliable basis for subsequent communication quality optimization and problem localization.

[0115] Further, determining the session-only label of the voice session data packet sample based on the call type label of each data stream sample includes:

[0116] If the call type label of the calling side uplink data stream sample, the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the first session single-pass label.

[0117] If the call type label of the calling side uplink data stream sample and the called side downlink data stream sample are both the single-pass label, and the call type label of the calling side downlink data stream sample and the called side uplink data stream sample are both the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the second session single-pass label.

[0118] If the call type label of the calling side downlink data stream sample and the called side uplink data stream sample are both the single-pass label, and the call type label of the calling side uplink data stream sample and the called side downlink data stream sample are both the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the third session single-pass label.

[0119] If the call type label of the calling side downlink data stream sample is the single-pass label, and the call type labels of the calling side uplink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the non-single-pass labels, then the session single-pass label of the voice session data packet sample is determined to be the fourth session single-pass label.

[0120] If the call type label of the called side downlink data stream sample is the single-pass label, and the call type labels of the calling side uplink data stream sample, the calling side downlink data stream sample, and the called side uplink data stream sample are all the non-single-pass labels, then the session single-pass label of the voice session data packet sample is determined to be the fifth session single-pass label.

[0121] If the call type labels of the calling side downlink data stream sample and the called side downlink data stream sample are both the single-pass label, and the call type labels of the calling side uplink data stream sample and the called side uplink data stream sample are both the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the sixth session single-pass label.

[0122] If the call type label of the calling side uplink data stream sample, the calling side downlink data stream sample, and the called side downlink data stream sample are all the single-pass label, and the call type label of the called side uplink data stream sample is the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the seventh session single-pass label.

[0123] If the call type label of the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the single-pass label, and the call type label of the calling side uplink data stream sample is the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the eighth session single-pass label.

[0124] If the call type label of the calling side uplink data stream sample, the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the ninth session single-pass label.

[0125] It should be noted that the session one-way label of the voice session data packet sample is obtained by combining the call type labels corresponding to the calling side uplink data stream sample, calling side downlink data stream sample, called side uplink data stream sample, and called side downlink data stream sample, respectively. The format of the session one-way label of the voice session data packet sample can be: MOX-MTX, where MO represents the calling side; MT represents the called side; X takes values ​​of 0, 1, 2, 3, where 0 represents neither uplink nor downlink is one-way, 1 represents uplink one-way and downlink is not one-way, 2 represents downlink one-way and uplink is not one-way, and 3 represents both uplink and downlink is one-way. Theoretically, there can be 16 combinations of session one-way labels in the MOX-MTX format. However, based on business logic, this embodiment of the invention determines 9 combinations of session one-way labels that conform to the business logic.

[0126] Specifically, if the call type labels of the calling side uplink data stream sample, the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all non-one-way labels, the voice one-way detection device determines that the session one-way label of the voice session data packet sample is the first session one-way label. Therefore, the first session one-way label can be represented as MO0-MT0.

[0127] Furthermore, if the call type labels of both the calling side uplink data stream sample and the called side downlink data stream sample are one-way labels, and the call type labels of both the calling side downlink data stream sample and the called side uplink data stream sample are non-one-way labels, then the voice one-way detection device determines that the session one-way label of the voice session data packet sample is the second session one-way label. Therefore, correspondingly, the second session one-way label can be represented as MO1-MT2.

[0128] Furthermore, if the call type labels of both the calling side downlink data stream sample and the called side uplink data stream sample are one-way labels, and the call type labels of both the calling side uplink data stream sample and the called side downlink data stream sample are non-one-way labels, then the voice one-way detection device determines that the session one-way label of the voice session data packet sample is a third session one-way label. Therefore, correspondingly, the third session one-way label can be represented as MO2-MT1.

[0129] Furthermore, if the call type label of the calling side downlink data stream sample is a one-way label, and the call type labels of the calling side uplink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all non-one-way labels, then the voice one-way detection device determines that the session one-way label of the voice session data packet sample is the fourth session one-way label. Therefore, correspondingly, the fourth session one-way label can be represented as MO2-MT0.

[0130] Furthermore, if the call type label of the called-side downlink data stream sample is a one-way label, and the call type labels of the calling-side uplink data stream sample, the calling-side downlink data stream sample, and the called-side uplink data stream sample are all non-one-way labels, then the voice one-way detection device determines that the session one-way label of the voice session data packet sample is the fifth session one-way label. Therefore, correspondingly, the fifth session one-way label can be represented as MO0-MT2.

[0131] Furthermore, if the call type labels of both the calling side downlink data stream sample and the called side downlink data stream sample are single-pass labels, and the call type labels of both the calling side uplink data stream sample and the called side uplink data stream sample are non-single-pass labels, then the voice single-pass detection device determines that the session single-pass label of the voice session data packet sample is the sixth session single-pass label. Therefore, correspondingly, the sixth session single-pass label can be represented as MO2-MT2.

[0132] Furthermore, if the call type labels of the calling side uplink data stream sample, the calling side downlink data stream sample, and the called side downlink data stream sample are all one-way labels, and the call type label of the called side uplink data stream sample is a non-one-way label, then the voice one-way detection device determines that the session one-way label of the voice session data packet sample is the seventh session one-way label. Therefore, correspondingly, the seventh session one-way label can be represented as MO3-MT2.

[0133] Furthermore, if the call type labels of the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all one-way labels, and the call type label of the calling side uplink data stream sample is a non-one-way label, then the voice one-way detection device determines that the session one-way label of the voice session data packet sample is the eighth session one-way label. Therefore, correspondingly, the eighth session one-way label can be represented as MO2-MT3.

[0134] Furthermore, if the call type labels of the calling side uplink data stream sample, the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all non-single-pass labels, the voice single-pass detection device determines that the session single-pass label of the voice session data packet sample is the ninth session single-pass label. Therefore, correspondingly, the ninth session single-pass label can be represented as MO3-MT3.

[0135] This invention, based on the call type labels corresponding to the calling side uplink data stream samples, calling side downlink data stream samples, called side uplink data stream samples, and called side downlink data stream samples, determines the session single-pass label of the voice session data packet samples. Then, based on the session single-pass label, model training is performed, realizing the model training by fusing multi-dimensional features and building a multi-feature fusion single-pass detection model. Therefore, the single-pass detection model can realize multi-dimensional single-pass detection results of calling side uplink, calling side downlink, called side uplink, and called side downlink in a session. The detection results can be used for further single-pass delimitation analysis, thereby improving the efficiency and reliability of the communication network.

[0136] Further, the step of training the boosting tree model based on the voice session data packet samples, the data type, the session one-way label, and the multi-segment data stream samples to obtain a first one-way detection model, a second one-way detection model, a third one-way detection model, and a fourth one-way detection model includes:

[0137] The voice conversation data packet sample belonging to the first type is identified as the first type of conversation data packet sample; based on the first type of conversation data packet sample and the conversation one-way label of the first type of conversation data packet sample, the boosting tree model is trained to obtain the first one-way detection model;

[0138] The voice conversation data packet samples belonging to the second type are identified as second type conversation data packet samples; based on the second type conversation data packet samples and the conversation one-way labels of the second type conversation data packet samples, the boosting tree model is trained to obtain the second one-way detection model;

[0139] The voice conversation data packet sample belonging to the third type is identified as the third type conversation data packet sample; the data stream sample in the multi-segment data stream sample of the third type conversation data packet sample that does not have missing RTP data packets is identified as the first target data stream sample; the boosting tree model is trained based on the conversation one-way label of the first target data stream sample and the third type conversation data packet sample to obtain the third one-way detection model;

[0140] The voice conversation data packet sample belonging to the fourth type is identified as the fourth type conversation data packet sample; the data stream sample in the multi-segment data stream sample of the fourth type conversation data packet sample that does not have missing RTP data packets is identified as the second target data stream sample; based on the conversation one-way label of the second target data stream sample and the fourth type conversation data packet sample, the lifting tree model is trained to obtain the fourth one-way detection model.

[0141] It should be noted that before training the single-pass detection model, three classification models were selected for algorithm selection experiments: XGBoost, K-Nearest Neighbors (KNN), and Support Vector Classification (SVC). Model accuracy and computational efficiency were used as evaluation indicators for the selection results. The accuracy of the models based on the XGBoost algorithm was all above 80%, and the time consumption was relatively small. Therefore, the XGBoost model was selected as the final single-pass detection model for training.

[0142] It should be further noted that, to avoid model underfitting, the ratio of samples carrying single-pass labels to samples carrying non-single-pass labels should be maintained at 1:1, and the ratio of various types of voice conversation data packets carrying single-pass labels should also be maintained at 1:1. Simultaneously, data augmentation methods based on Gaussian noise can be used to construct the required number of samples for model training, thereby enhancing the sample data.

[0143] Specifically, the voice one-way detection device identifies voice conversation data packet samples belonging to the first type as first type conversation data packet samples. Furthermore, the voice one-way detection device trains the XGBoost model based on the first type conversation data packet samples and their conversation one-way labels to obtain the first one-way detection model.

[0144] Furthermore, the voice one-way detection device identifies voice session data packet samples belonging to the second type as second type session data packet samples. Further, the voice one-way detection device trains the XGBoost model based on the second type session data packet samples and their session one-way labels to obtain the second one-way detection model.

[0145] It should be noted that the RTP data packets of the voice conversation data packet samples belonging to the first or second type are complete, and the voice conversation data packet samples can be directly used as training data for model training.

[0146] Furthermore, the voice one-way detection device identifies voice session data packet samples belonging to the third type as third type session data packet samples. Furthermore, the voice one-way detection device identifies data stream samples in the multi-segment data stream samples of the third type session data packet samples that do not contain missing RTP data packets as the first target data stream samples.

[0147] Furthermore, the voice one-way detection device trains the XGBoost model based on the session one-way labels of the first target data stream sample and the third type of session data packet sample to obtain the third one-way detection model.

[0148] Furthermore, the voice one-way detection device identifies voice session data packet samples belonging to the fourth type as fourth type session data packet samples. Furthermore, the voice one-way detection device identifies data stream samples in the multi-segment data stream samples of the fourth type session data packet samples that do not contain missing RTP data packets as the second target data stream samples.

[0149] Furthermore, the voice one-way detection device trains the XGBoost model based on the session one-way labels of the second target data stream samples and the fourth type of session data packet samples to obtain the fourth one-way detection model.

[0150] It should be noted that the RTP packets in the voice session data packet samples belonging to the third or fourth type are incomplete. Therefore, it is necessary to identify data stream samples in the multi-segment data stream samples that do not have missing RTP packets and use them as the training set for model training. For data stream samples in the multi-segment data stream samples that have missing RTP packets, their session one-way label is set to the default value of 0 and they are not included in the model training. However, the label results of this type of data stream can be used to determine the label of its corresponding voice session data packet sample.

[0151] It should be further explained that when splitting a voice session data packet sample into multiple data stream samples for model training, since there may be many sample data, each voice session data packet sample will have a unique number. Therefore, each of the multiple data stream samples obtained from splitting a voice session data packet sample will correspond to the unique number of that voice session data packet sample. This allows the multiple data stream samples belonging to a single voice session data packet sample to be aggregated during model training to obtain the session single-pass label of that voice session data packet sample.

[0152] In this invention, different models are trained to adapt to different data types, making the models more closely match the characteristics of specific data types and improving the personalization and accuracy of the models. Therefore, voice conversation data packets with varying degrees of missing fields can achieve single-pass detection by selecting the appropriate model, which greatly improves the coverage of voice single-pass detection and thus improves the accuracy of voice single-pass detection.

[0153] Furthermore, the present invention also provides a voice one-way detection device.

[0154] Reference Figure 2 , Figure 2 This is a schematic diagram of the structure of the voice one-way detection device provided by the present invention.

[0155] The voice one-way detection device includes:

[0156] Module 210 is used to acquire voice session data packets;

[0157] The splitting module 220 is used to split the voice session data packets according to the data flow direction to obtain multiple data streams;

[0158] The first determining module 230 is used to determine the missing status of Real-Time Transport Protocol (RTP) data packets and Real-Time Transport Control Protocol (RTCP) data packets in each data stream.

[0159] The second determining module 240 is used to determine the data type of the voice session data packet based on the missing RTP data packets and the missing RTCP data packets of each data stream.

[0160] The single-pass detection module 250 is used to perform single-pass detection on the voice session data packet based on the data type, and obtain the single-pass detection result.

[0161] The voice one-way detection device provided by this invention splits the acquired voice session data packets according to the data flow direction to obtain multiple data streams. It then fuses multi-dimensional features for voice one-way detection. Furthermore, based on the missing data packets of the Real-Time Transmission Protocol (RTP) and Real-Time Transmission Control Protocol (RTC) in each data stream, it determines the data type of the voice session data packets, thus classifying voice session data packets with different missing data conditions. Finally, it performs one-way detection based on these different missing data packets to obtain the one-way detection results. This reduces the impact of missing data in some fields on the detection of the entire voice session data packet, ensuring the coverage of voice one-way detection and thereby improving the accuracy of voice one-way detection.

[0162] Furthermore, the second determining module 240 also includes:

[0163] If there are no missing RTP packets in all data streams and no missing RTCP packets in all data streams, then the data type of the voice session data packet is determined to be the first type.

[0164] If all RTP packets in all data streams are complete and at least one RTCP packet in a data stream is missing, then the data type of the voice session data packet is determined to be the second type.

[0165] If at least one data stream has missing RTP packets, and all data streams have no missing RTCP packets, then the data type of the voice session data packets is determined to be the third type.

[0166] If at least one segment of the RTP data packets is missing, and at least one segment of the RTCP data packets is missing, then the data type of the voice session data packets is determined to be the fourth type.

[0167] Furthermore, the single-pass detection module 250 also includes:

[0168] If the data type is the first type, the voice conversation data packet is input into the first single-pass detection model to obtain the first single-pass detection result output by the first single-pass detection model; the first single-pass detection model is obtained by training the model based on the first type of conversation data packet sample and the conversation single-pass label of the first type of conversation data packet sample.

[0169] If the data type is the second type, the voice conversation data packet is input into the second single-pass detection model to obtain the second single-pass detection result output by the second single-pass detection model; the second single-pass detection model is obtained by training the model based on the second type of conversation data packet sample and the conversation single-pass label of the second type of conversation data packet sample.

[0170] If the data type is the third type, then the multi-segment data stream is input into the third single-pass detection model to obtain the third single-pass detection result output by the third single-pass detection model; the third single-pass detection model is obtained by training the model based on the multi-segment data stream samples of the third type session data packet samples and the session single-pass labels of the third type session data packet samples;

[0171] If the data type is the fourth type, then the multi-segment data stream is input into the fourth single-pass detection model to obtain the fourth single-pass detection result output by the fourth single-pass detection model; the fourth single-pass detection model is obtained by training the model based on the multi-segment data stream samples of the fourth type session data packet samples and the session single-pass labels of the fourth type session data packet samples.

[0172] Furthermore, the voice one-way detection device also includes:

[0173] Obtain voice session data packet samples;

[0174] The voice session data packet sample is split according to the data flow direction to obtain multiple data flow samples;

[0175] Determine the call type label for each data stream sample; the call type label includes a one-way label and a non-one-way label;

[0176] Based on the call type label of each data stream sample, determine the session one-way label of the voice session data packet sample;

[0177] Based on the missing RTP packets and missing RTCP packets in each data stream sample, the data type of the voice session data packet sample is determined.

[0178] Based on the voice session data packet samples, the data type, the session one-way label, and the multi-segment data stream samples, the boosting tree model is trained to obtain the first one-way detection model, the second one-way detection model, the third one-way detection model, and the fourth one-way detection model, respectively.

[0179] Furthermore, the voice one-way detection device also includes:

[0180] If the call type label of the calling side uplink data stream sample, the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the first session single-pass label.

[0181] If the call type label of the calling side uplink data stream sample and the called side downlink data stream sample are both the single-pass label, and the call type label of the calling side downlink data stream sample and the called side uplink data stream sample are both the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the second session single-pass label.

[0182] If the call type label of the calling side downlink data stream sample and the called side uplink data stream sample are both the single-pass label, and the call type label of the calling side uplink data stream sample and the called side downlink data stream sample are both the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the third session single-pass label.

[0183] If the call type label of the calling side downlink data stream sample is the single-pass label, and the call type labels of the calling side uplink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the non-single-pass labels, then the session single-pass label of the voice session data packet sample is determined to be the fourth session single-pass label.

[0184] If the call type label of the called side downlink data stream sample is the single-pass label, and the call type labels of the calling side uplink data stream sample, the calling side downlink data stream sample, and the called side uplink data stream sample are all the non-single-pass labels, then the session single-pass label of the voice session data packet sample is determined to be the fifth session single-pass label.

[0185] If the call type labels of the calling side downlink data stream sample and the called side downlink data stream sample are both the single-pass label, and the call type labels of the calling side uplink data stream sample and the called side uplink data stream sample are both the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the sixth session single-pass label.

[0186] If the call type label of the calling side uplink data stream sample, the calling side downlink data stream sample, and the called side downlink data stream sample are all the single-pass label, and the call type label of the called side uplink data stream sample is the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the seventh session single-pass label.

[0187] If the call type label of the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the single-pass label, and the call type label of the calling side uplink data stream sample is the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the eighth session single-pass label.

[0188] If the call type label of the calling side uplink data stream sample, the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the ninth session single-pass label.

[0189] Furthermore, the voice one-way detection device also includes:

[0190] The voice conversation data packet sample belonging to the first type is identified as the first type of conversation data packet sample; based on the first type of conversation data packet sample and the conversation one-way label of the first type of conversation data packet sample, the boosting tree model is trained to obtain the first one-way detection model;

[0191] The voice conversation data packet samples belonging to the second type are identified as second type conversation data packet samples; based on the second type conversation data packet samples and the conversation one-way labels of the second type conversation data packet samples, the boosting tree model is trained to obtain the second one-way detection model;

[0192] The voice conversation data packet sample belonging to the third type is identified as the third type conversation data packet sample; the data stream sample in the multi-segment data stream sample of the third type conversation data packet sample that does not have missing RTP data packets is identified as the first target data stream sample; the boosting tree model is trained based on the conversation one-way label of the first target data stream sample and the third type conversation data packet sample to obtain the third one-way detection model;

[0193] The voice conversation data packet sample belonging to the fourth type is identified as the fourth type conversation data packet sample; the data stream sample in the multi-segment data stream sample of the fourth type conversation data packet sample that does not have missing RTP data packets is identified as the second target data stream sample; based on the conversation one-way label of the second target data stream sample and the fourth type conversation data packet sample, the lifting tree model is trained to obtain the fourth one-way detection model.

[0194] It should be noted that the voice one-way detection device provided by the present invention can execute the voice one-way detection method described in any of the above embodiments during specific operation, which will not be elaborated in this embodiment.

[0195] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 3 As shown, the electronic device may include a processor 310, a communications interface 320, a memory 330, and a communication bus 340. The processor 310, communications interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions in the memory 330 to execute a voice one-way detection method. This method includes: acquiring voice session data packets; splitting the voice session data packets according to the data flow direction to obtain multiple data streams; determining the missing status of Real-Time Transport Protocol (RTP) data packets and Real-Time Transport Control Protocol (RTCP) data packets in each data stream; determining the data type of the voice session data packets based on the missing RTP and RTCP data packets in each data stream; and performing one-way detection on the voice session data packets based on the data type to obtain a one-way detection result.

[0196] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0197] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, and when the program instructions are executed by a computer, the computer is able to execute the voice one-way detection method provided in the above embodiments, the method including: acquiring voice session data packets; splitting the voice session data packets according to the data flow direction to obtain multiple data streams; determining the missing status of Real-Time Transport Protocol (RTP) data packets and Real-Time Transport Control Protocol (RTCP) data packets in each data stream; determining the data type of the voice session data packets based on the missing status of RTP data packets and RTCP data packets in each data stream; and performing one-way detection on the voice session data packets based on the data type to obtain a one-way detection result.

[0198] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the voice one-way detection method provided in the above embodiments. The method includes: acquiring voice session data packets; splitting the voice session data packets according to the data flow direction to obtain multiple data streams; determining the missing status of Real-Time Transport Protocol (RTP) data packets and Real-Time Transport Control Protocol (RTCP) data packets in each data stream; determining the data type of the voice session data packets based on the missing status of RTP and RTCP data packets in each data stream; and performing one-way detection on the voice session data packets based on the data type to obtain a one-way detection result.

[0199] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0200] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0201] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting one-way speech, characterized in that, include: Obtain voice session data packets; The voice session data packets are split according to the data flow direction to obtain multiple data streams; Determine the missing Real-Time Transport Protocol (RTP) packets and Real-Time Transport Control Protocol (RTCP) packets for each data stream. Based on the missing RTP packets and missing RTCP packets in each data stream, the data type of the voice session data packets is determined. Based on the data type, one-way detection is performed on the voice session data packets to obtain the one-way detection result.

2. The voice one-way detection method according to claim 1, characterized in that, The determination of the data type of the voice session data packets based on the missing RTP and RTCP data packets of each data stream includes: If there are no missing RTP packets in all data streams and no missing RTCP packets in all data streams, then the data type of the voice session data packet is determined to be the first type. If all RTP packets in all data streams are complete and at least one RTCP packet in a data stream is missing, then the data type of the voice session data packet is determined to be the second type. If at least one data stream has missing RTP packets, and all data streams have no missing RTCP packets, then the data type of the voice session data packets is determined to be the third type. If at least one segment of the RTP data packets is missing, and at least one segment of the RTCP data packets is missing, then the data type of the voice session data packets is determined to be the fourth type.

3. The voice one-way detection method according to claim 2, characterized in that, The step of performing one-way detection on the voice session data packets based on the data type to obtain the one-way detection result includes: If the data type is the first type, the voice conversation data packet is input into the first single-pass detection model to obtain the first single-pass detection result output by the first single-pass detection model; the first single-pass detection model is obtained by training the model based on the first type of conversation data packet sample and the conversation single-pass label of the first type of conversation data packet sample. If the data type is the second type, the voice conversation data packet is input into the second single-pass detection model to obtain the second single-pass detection result output by the second single-pass detection model; the second single-pass detection model is obtained by training the model based on the second type of conversation data packet sample and the conversation single-pass label of the second type of conversation data packet sample. If the data type is the third type, then the multi-segment data stream is input into the third single-pass detection model to obtain the third single-pass detection result output by the third single-pass detection model; the third single-pass detection model is obtained by training the model based on the multi-segment data stream samples of the third type session data packet samples and the session single-pass labels of the third type session data packet samples; If the data type is the fourth type, then the multi-segment data stream is input into the fourth single-pass detection model to obtain the fourth single-pass detection result output by the fourth single-pass detection model; the fourth single-pass detection model is obtained by training the model based on the multi-segment data stream samples of the fourth type session data packet samples and the session single-pass labels of the fourth type session data packet samples.

4. The voice one-way detection method according to claim 1, characterized in that, The multiple data streams include the calling side uplink data stream, the calling side downlink data stream, the called side uplink data stream, and the called side downlink data stream.

5. The voice one-way detection method according to claim 3, characterized in that, The first single-pass detection model, the second single-pass detection model, the third single-pass detection model, and the fourth single-pass detection model are obtained through training in the following manner: Obtain voice session data packet samples; The voice session data packet sample is split according to the data flow direction to obtain multiple data flow samples; Determine the call type label for each data stream sample; the call type label includes a one-way label and a non-one-way label; Based on the call type label of each data stream sample, determine the session one-way label of the voice session data packet sample; Based on the missing RTP packets and missing RTCP packets in each data stream sample, the data type of the voice session data packet sample is determined. Based on the voice session data packet samples, the data type, the session one-way label, and the multi-segment data stream samples, the boosting tree model is trained to obtain the first one-way detection model, the second one-way detection model, the third one-way detection model, and the fourth one-way detection model, respectively.

6. The voice one-way detection method according to claim 5, characterized in that, The multi-segment data stream samples include the calling side uplink data stream sample, the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample. The process of determining the session-only label of the voice session data packet sample based on the call type label of each data stream sample includes: If the call type label of the calling side uplink data stream sample, the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the first session single-pass label. If the call type label of the calling side uplink data stream sample and the called side downlink data stream sample are both the single-pass label, and the call type label of the calling side downlink data stream sample and the called side uplink data stream sample are both the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the second session single-pass label. If the call type label of the calling side downlink data stream sample and the called side uplink data stream sample are both the single-pass label, and the call type label of the calling side uplink data stream sample and the called side downlink data stream sample are both the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the third session single-pass label. If the call type label of the calling side downlink data stream sample is the single-pass label, and the call type labels of the calling side uplink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the non-single-pass labels, then the session single-pass label of the voice session data packet sample is determined to be the fourth session single-pass label. If the call type label of the called side downlink data stream sample is the single-pass label, and the call type labels of the calling side uplink data stream sample, the calling side downlink data stream sample, and the called side uplink data stream sample are all the non-single-pass labels, then the session single-pass label of the voice session data packet sample is determined to be the fifth session single-pass label. If the call type labels of the calling side downlink data stream sample and the called side downlink data stream sample are both the single-pass label, and the call type labels of the calling side uplink data stream sample and the called side uplink data stream sample are both the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the sixth session single-pass label. If the call type label of the calling side uplink data stream sample, the calling side downlink data stream sample, and the called side downlink data stream sample are all the single-pass label, and the call type label of the called side uplink data stream sample is the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the seventh session single-pass label. If the call type label of the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the single-pass label, and the call type label of the calling side uplink data stream sample is the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the eighth session single-pass label. If the call type label of the calling side uplink data stream sample, the calling side downlink data stream sample, the called side uplink data stream sample, and the called side downlink data stream sample are all the non-single-pass label, then the session single-pass label of the voice session data packet sample is determined to be the ninth session single-pass label.

7. The voice one-way detection method according to claim 6, characterized in that, The boosting tree model is trained based on the voice session data packet samples, the data type, the session one-way label, and the multi-segment data stream samples to obtain a first one-way detection model, a second one-way detection model, a third one-way detection model, and a fourth one-way detection model, including: The voice conversation data packet sample belonging to the first type is identified as the first type of conversation data packet sample; based on the first type of conversation data packet sample and the conversation one-way label of the first type of conversation data packet sample, the boosting tree model is trained to obtain the first one-way detection model; The voice conversation data packet samples belonging to the second type are identified as second type conversation data packet samples; based on the second type conversation data packet samples and the conversation one-way labels of the second type conversation data packet samples, the boosting tree model is trained to obtain the second one-way detection model; The voice conversation data packet sample belonging to the third type is identified as the third type conversation data packet sample; the data stream sample in the multi-segment data stream sample of the third type conversation data packet sample that does not have missing RTP data packets is identified as the first target data stream sample; the boosting tree model is trained based on the conversation one-way label of the first target data stream sample and the third type conversation data packet sample to obtain the third one-way detection model; The voice conversation data packet sample belonging to the fourth type is identified as the fourth type conversation data packet sample; the data stream sample in the multi-segment data stream sample of the fourth type conversation data packet sample that does not have missing RTP data packets is identified as the second target data stream sample; based on the conversation one-way label of the second target data stream sample and the fourth type conversation data packet sample, the lifting tree model is trained to obtain the fourth one-way detection model.

8. A voice one-way detection device, characterized in that, include: The acquisition module is used to acquire voice session data packets; The splitting module is used to split the voice session data packets according to the data flow direction to obtain multiple data streams; The first determining module is used to determine the missing status of Real-Time Transport Protocol (RTP) data packets and Real-Time Transport Control Protocol (RTCP) data packets in each data stream. The second determining module is used to determine the data type of the voice session data packet based on the missing RTP data packets and the missing RTCP data packets of each data stream. The single-pass detection module is used to perform single-pass detection on the voice session data packets based on the data type, and obtain the single-pass detection result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the voice one-way detection method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the voice one-way detection method as described in any one of claims 1 to 7.

11. A computer program product, the computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the voice one-way detection method as described in any one of claims 1 to 7.

Images