Speech generation method and device based on intelligent outbound call, equipment and storage medium

Abstract

The embodiment of the invention provides a verbal skill generation method and device based on an intelligent outbound call, equipment and a storage medium, and belongs to the technical field of financial science and technology. The method comprises the following steps: acquiring a target scene task of a target object and historical call-out voice of the target object in a historical scene task; performing voice extraction on the historical outbound voice and the historical node verbal skill text based on a voice extraction model to obtain historical object voice; performing intention recognition on the historical object voice based on an intention recognition model to obtain a node reply intention category and category matching data of the historical task node; determining a node confirmation state of the target object for the historical task node based on the node reply intention category and the category matching data; and generating a target outbound call based on the node confirmation state and the scene task node. According to the embodiment of the invention, the matching accuracy of the object intention and the talk operation for intelligent outbound calling can be improved, and a more accurate outbound talk operation is generated.

Description

Technical Field

[0001] The present application relates to the field of financial technology, and in particular to a method and device, equipment, and storage medium for generating speech based on intelligent outbound calls. Background Art

[0002] Smart outbound calling is a service that uses artificial intelligence technology to automatically make calls and communicate. For example, in the insurance scenario of financial technology, smart outbound calling can usually be used to remind the reminder object of key nodes, and the process of reminding key nodes can include identity confirmation of the reminder object, reminders of specific matters, and appointments for the next return visit. In addition, this type of smart outbound calling method usually pre-sets a combination of words, and broadcasts the corresponding words by identifying the answer or question of the reminder object, thereby replacing the work of manual customer service outbound calls to a certain extent to reduce labor costs.

[0003] However, since the process of intelligent outbound calls to remind key nodes is generally long, it may be necessary to remind the reminder object multiple times before the node task expires. The current method of pre-setting the script combination easily causes the reminder object to receive multiple outbound calls containing the same question type script, resulting in a low information reach rate for the outbound call task and affecting the customer experience. Therefore, how to improve the accuracy of matching the object's intention with the script used for intelligent outbound calls, generate more accurate outbound call scripts, and thus improve the information reach rate of outbound call tasks has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose a method and device, equipment, and storage medium for generating speech based on intelligent outbound calls, aiming to improve the accuracy of matching the object intention with the speech used for intelligent outbound calls, generate more accurate outbound call speech, and thus improve the information reach rate of outbound call tasks.

[0005] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a method for generating speech based on intelligent outbound calling, the method comprising:

[0006] Acquire a target scenario task of a target object, wherein the target scenario task includes a scenario task node;

[0007] Obtain the historical outbound voice of the target object in the historical scenario task, wherein the historical scenario task includes a historical task node and a historical node speech text of the historical task node;

[0008] Performing voice extraction on the historical outbound call voice and the historical node speech text based on a preset voice extraction model to obtain a historical object voice, where the historical object voice is the voice of the target object in replying to the historical node speech text in the historical outbound call conversation;

[0009] Performing intent recognition on the historical object speech based on the intent recognition model to obtain the node reply intent category and category matching data of the historical task node, wherein the category matching data is used to indicate the distance between the node reply intent category and the speech text of the historical object speech;

[0010] Determine the node confirmation status of the target object for the historical task node based on the node reply intention category and the category matching data;

[0011] Generate target outbound call speech based on the node confirmation status and the scenario task node.

[0012] In some embodiments, generating a target outbound call speech based on the node confirmation state and the scenario task node includes:

[0013] Divide the scene task nodes into nodes based on the node confirmation status to obtain target confirmed nodes and target unconfirmed nodes;

[0014] Obtaining node confirmation data of the target object for the target confirmed node;

[0015] Based on the speech generation model, speech generation is performed on the target confirmed node and the node confirmation data to obtain the outbound call opening speech;

[0016] Generate speech for the target unconfirmed node based on the speech generation model to obtain the speech for the outbound unconfirmed node;

[0017] The target outbound call speech is obtained by combining the speech based on the outbound call opening speech and the outbound call unconfirmed node speech.

[0018] In some embodiments, dividing the scene task nodes based on the node confirmation status to obtain target confirmed nodes and target unconfirmed nodes includes:

[0019] Selecting a historical confirmed node from the historical task nodes based on the node confirmation status;

[0020] Perform node matching on the historically confirmed nodes and the scenario task nodes to obtain a node matching result;

[0021] The target confirmed node and the target unconfirmed node are determined from the scenario task nodes based on the node matching result.

[0022] In some embodiments, the method further comprises:

[0023] Obtaining the task priority of the target scenario task;

[0024] Matching the target scenario tasks with interaction modes based on the task priorities to determine a target interaction mode, where the target interaction mode is used to indicate a communication mode used to confirm the target task node to the target object;

[0025] Based on the target outbound calling speech and the target interaction mode, an intelligent outbound call is made to the target object.

[0026] In some embodiments, matching the target scene task with an interaction mode based on the task priority to determine the target interaction mode includes:

[0027] Obtaining a pre-assigned interaction mode for the target scenario task;

[0028] Performing saturation division on the preset interaction benchmark number of the pre-allocated interaction mode based on a preset saturation division function to obtain saturation division intervals, wherein the saturation division intervals include a first saturation interval and a second saturation interval, and the saturation degree of the first saturation interval is higher than the saturation degree of the second saturation interval;

[0029] Performing interval matching on the number of interaction objects to be allocated of the pre-allocated interaction mode based on the saturation degree sub-interval, and determining a matching interval of the pre-allocated interaction mode;

[0030] If the matching interval is the first saturation interval, the pre-allocated interaction mode is adjusted based on the task priority to obtain the target interaction mode.

[0031] In some embodiments, the performing voice extraction on the historical outbound voice and the historical node speech text based on a preset voice extraction model to obtain the historical object voice includes:

[0032] Performing noise reduction processing on the historical outbound call voice to obtain noise-reduced outbound call voice;

[0033] Performing speech encoding on the noise-reduced outbound voice to obtain outbound voice features;

[0034] Performing voice feature separation on the outbound voice feature to obtain voice features of an outbound call recipient, wherein the voice features of the outbound call recipient are used to indicate different speaking recipients in the historical outbound voice;

[0035] Decoding the voice features of the outbound call recipient to obtain a voice segment of the outbound call recipient;

[0036] Selecting a reply object voice segment from the outbound call object voice segment based on the historical node speech text, wherein the reply object voice segment is a voice segment in which the target object replies to the historical node speech text in the historical outbound call conversation;

[0037] The reply object voice segments are integrated to obtain the historical object voice.

[0038] In some embodiments, the performing intent recognition on the historical object speech based on the intent recognition model to obtain the node reply intent category and category matching data of the historical task node includes:

[0039] Obtaining a node verification category of the historical task node corresponding to the reply object voice segment, wherein the node verification category includes an intent recognition category and a labeling category;

[0040] If the node verification category is the intention recognition category, perform intention recognition on the reply object voice segment based on the intention recognition model to obtain the node reply intention category and the category matching data;

[0041] If the node verification category is the marked category, the preset category is determined to be the node reply intention category of the reply object voice segment, and the preset matching data is determined to be the category matching data of the reply object voice segment.

[0042] To achieve the above-mentioned purpose, the second aspect of the embodiment of the present application proposes a speech generation device based on intelligent outbound calling, the device comprising:

[0043] A first acquisition module, used to acquire a target scene task of a target object, wherein the target scene task includes a scene task node;

[0044] The second acquisition module is used to acquire the historical outbound voice of the target object in the historical scenario task, wherein the historical scenario task includes a historical task node and a historical node speech text of the historical task node;

[0045] An extraction module, for performing speech extraction on the historical outbound call speech and the historical node speech text based on a preset speech extraction model to obtain a historical object speech, wherein the historical object speech is the speech of the target object in replying to the historical node speech text in the historical outbound call dialogue;

[0046] A recognition module, used to perform intent recognition on the historical object speech based on an intent recognition model, and obtain a node reply intent category and category matching data of the historical task node, wherein the category matching data is used to indicate a distance between the node reply intent category and the speech text of the historical object speech;

[0047] A determination module, configured to determine a node confirmation status of the target object for the historical task node based on the node reply intention category and the category matching data;

[0048] A generation module is used to generate target outbound call speech based on the node confirmation status and the scenario task node.

[0049] To achieve the above objectives, a third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.

[0050] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0051] The present application proposes a method and device, equipment, and storage medium for generating speech based on intelligent outbound calls, which obtains a target scenario task of a target object, wherein the target scenario task includes a scenario task node; obtains the historical outbound voice of the target object in a historical scenario task, wherein the historical scenario task includes a historical task node and a historical node speech text of the historical task node; further, based on a preset speech extraction model, speech extraction is performed on the historical outbound voice and the historical node speech text to obtain a historical object voice, wherein the historical object voice is the voice of the target object in replying to the historical node speech text in the historical outbound call conversation; further, based on an intention recognition model, intent recognition is performed on the historical object voice to obtain a node reply intent category and category matching data of the historical task node, wherein the category matching data is used to indicate the distance between the node reply intent category and the speech text of the historical object voice; further, based on the node reply intent category and the category matching data, the node confirmation status of the target object for the historical task node is determined; further, based on the node reply intent category and the category matching data, the target outbound speech is generated based on the node confirmation status and the scenario task node. Compared with the related art method of using a combination of pre-set fixed words, it is easy to cause the reminder object to receive multiple outbound calls containing the same question type words, resulting in a low information reach rate for the outbound call task. The embodiment of the present application will determine the node confirmation status of the historical task node from the historical object voice by combining voice extraction and intent recognition, thereby generating a target outbound call wording that is more suitable for the current scenario task node. In this way, it is possible to avoid the target object from receiving multiple outbound calls containing the same question type words, improve the accuracy of matching the object's intention with the words used for intelligent outbound calls, generate more accurate outbound call words, and thus improve the information reach rate of the outbound call task. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a flow chart of a method for generating speech based on intelligent outbound calling provided in an embodiment of the present application;

[0053] Figure 2 yes Figure 1A flowchart of step S130 in FIG.

[0054] Figure 3 yes Figure 1 A flowchart of step S140 in FIG.

[0055] Figure 4 yes Figure 1 A flowchart of step S160 in FIG.

[0056] Figure 5 yes Figure 4 A flowchart of step S410 in FIG.

[0057] Figure 6 is another flow chart of the method for generating speech based on intelligent outbound calling provided in an embodiment of the present application;

[0058] Figure 7 yes Figure 6 A flowchart of step S620 in FIG.

[0059] Figure 8 It is a structural schematic diagram of a speech generation device based on intelligent outbound calling provided in an embodiment of the present application;

[0060] Fig. 9 It is a hardware structure diagram of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0062] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0063] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0064] First, some nouns involved in this application are analyzed:

[0065] Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0066] Natural language processing (NLP): NLP uses computers to process, understand and apply human languages ​​(such as Chinese, English, etc.). NLP is a branch of artificial intelligence and an interdisciplinary subject between computer science and linguistics. It is often referred to as computational linguistics. Natural language processing includes grammatical analysis, semantic analysis, and text understanding. Natural language processing is often used in technical fields such as machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, information intent recognition, information extraction and filtering, text classification and clustering, public opinion analysis and opinion mining. It involves data mining related to language processing, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, and linguistic research related to language computing.

[0067] Intent Recognition: refers to the process of using the system to identify the intent or purpose behind the user's input data such as sentences, text, and images.

[0068] Smart outbound calling is a service that uses artificial intelligence technology to automatically make calls and communicate. For example, in the insurance scenario of financial technology, smart outbound calling can usually be used to remind the reminder object of key nodes (such as insurance payment due, claim information submission, etc.), and the process of reminding key nodes can include identity confirmation of the reminder object, reminders of specific matters, and appointments for the next return visit. In addition, this type of smart outbound calling method usually pre-sets a combination of words, and broadcasts the corresponding words by identifying the reminder object's answer or question, thereby replacing the work of manual customer service outbound calls to a certain extent to reduce labor costs.

[0069] However, since the process of intelligent outbound calls to remind key nodes is generally long, it may be necessary to remind the reminder object multiple times before the node task expires. The current method of pre-setting the script combination easily causes the reminder object to receive multiple outbound calls containing the same question type script, resulting in a low information reach rate for the outbound call task and affecting the customer experience. Therefore, how to improve the accuracy of matching the object's intention with the script used for intelligent outbound calls, generate more accurate outbound call scripts, and thus improve the information reach rate of outbound call tasks has become a technical problem that needs to be solved urgently.

[0070] Based on this, the embodiments of the present application provide a method and device, equipment, and storage medium for generating speech based on intelligent outbound calls, aiming to improve the accuracy of matching the object's intention with the speech used for intelligent outbound calls, generate more accurate outbound call speech, and thus improve the information reach rate of outbound call tasks.

[0071] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0072] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0073] The method for generating speech based on intelligent outbound calls provided in the embodiment of the present application relates to the field of artificial intelligence technology. The method for generating speech based on intelligent outbound calls provided in the embodiment of the present application can be applied to a terminal, can be applied to a server side, or can be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the method for generating speech based on intelligent outbound calls, etc., but is not limited to the above forms.

[0074] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0075] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on the data related to the identity or characteristics of the object, such as the object's outbound voice, the object's historical data, and the object's insurance data, the object's permission or consent will be obtained first, and the collection, use and processing of these data will comply with relevant laws, regulations and standards. In addition, when the embodiment of the present application needs to obtain the sensitive personal information of the object, the object's separate permission or separate consent will be obtained through a pop-up window or jump to a confirmation page. After clearly obtaining the object's separate permission or separate consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0076] See also Figure 1 , Figure 1 is an optional flow chart of the method for generating speech based on intelligent outbound calls provided in an embodiment of the present application. In some embodiments of the present application, Figure 1 The method may specifically include but is not limited to steps S110 to S160.

[0077] Step S110, obtaining a target scene task of a target object;

[0078] Step S120, obtaining the historical outbound voice of the target object in the historical scenario task;

[0079] Step S130, performing voice extraction on the historical outbound voice and the historical node speech text based on a preset voice extraction model to obtain the historical object voice;

[0080] Step S140, performing intent recognition on the historical object speech based on the intent recognition model, and obtaining the node reply intent category and category matching data of the historical task node;

[0081] Step S150, determining the node confirmation status of the target object for the historical task node based on the node reply intention category and category matching data;

[0082] Step S160, generating target outbound call scripts based on the node confirmation status and scenario task nodes.

[0083] In step S110 of some embodiments, the target object refers to the object to be called out, that is, the object to be confirmed by the node through the set interactive method. The target scenario task refers to the task of the business scenario that the outbound call system needs to interact with the target object to complete, for example, the renewal payment reminder task, renewal business recommendation task, etc. in the insurance field. The target scenario task may include multiple scenario task nodes, and the outbound call system may complete the confirmation of the target scenario task by setting a series of predefined scenario task nodes. For example, the target scenario task is a renewal payment reminder task in the insurance field, and the scenario task nodes here may include identity confirmation, payment reminder, appointment for the next return visit, etc. In this way, there may be a sequence relationship, priority relationship, etc. between the scenario task nodes.

[0084] It should be noted that in intelligent outbound calls, each scenario task node can correspond to a specific dialogue target or question, that is, the configured outbound call script, which is used to generate the corresponding voice to confirm with the target object. Among them, the target scenario tasks and scenario task nodes can be pre-configured to achieve differentiated management of outbound call tasks for different business scenarios. And because each scenario task node may be different, the scope of intent recognition for different target scenario tasks will be different.

[0085] In step S120 of some embodiments, the historical scenario task refers to a scenario task that has been voice communicated with the target object through a set interactive method. The historical scenario task includes a historical task node and the historical node speech text of the historical task node. The historical task node refers to the task node that has been conveyed during the historical outbound call process, and the historical task node and the scenario task node can be the same or different, depending on whether the historical task node has completed the confirmation or whether the node configured by the target scenario task requires multiple confirmations. The historical node speech text refers to the pre-set speech text for confirming the historical task node to the target object. For example, if the historical task node is an identity confirmation, the corresponding historical node speech text can be "Excuse me, are you xxx?" The historical outbound call voice is the communication voice generated by the outbound call system after intelligent communication with the target object based on the historical node speech text.

[0086] In step S130 of some embodiments, the speech extraction model refers to a model for extracting speech of different objects from input speech. The speech extraction model can be trained based on a machine learning model, a neural network model, or the like. For example, the speech extraction model can be a model constructed based on the structure of a time-domain audio separation network (TasNet), which can achieve human voice separation while taking into account efficiency and performance, so as to obtain a human voice segment containing only the answer of the target object.

[0087] It should be noted that the historical object voice is the voice of the target object in replying to the historical node speech text in the historical outbound call dialogue. Among them, the target object in the historical object voice may have replied to the voice questions corresponding to all the historical node speech texts, or may have replied to the voice questions corresponding to some of the historical node speech texts, without specific limitation. In this way, by obtaining the reply voice of the target object in the historical outbound call voice, it is possible to analyze how the target object responded to a specific speech in the historical outbound call dialogue, thereby generating a speech that is more suitable for the target object, which is convenient for use in the next outbound call interaction.

[0088] See also Figure 2 , Figure 2 It is a specific flow chart of step S130 provided in an embodiment of the present application. In some embodiments of the present application, step S130 may specifically include but is not limited to steps S210 to S260.

[0089] Step S210, performing noise reduction processing on the historical outbound call voice to obtain noise-reduced outbound call voice;

[0090] Step S220, performing speech encoding on the noise-reduced outbound voice to obtain outbound voice features;

[0091] Step S230, performing voice feature separation on the outbound voice feature to obtain the voice feature of the outbound call recipient;

[0092] Step S240, decoding the voice features of the outbound call recipient to obtain a voice segment of the outbound call recipient;

[0093] Step S250, selecting a reply object voice segment from the outbound call object voice segment based on the historical node speech text;

[0094] Step S260, integrating the reply object speech segments to obtain the historical object speech.

[0095] In step S210 of some embodiments, considering that there may be some noise or background sound in the acquired historical outbound voice, in order to reduce the interference with the subsequent key intent recognition, the present application may first pre-process the historical outbound voice, and the pre-processing at this time may include voice cleaning, noise reduction processing, voice enhancement, etc., without limitation. Noise reduction processing (such as filters and spectral subtraction) can effectively remove background noise and interference sound, thereby obtaining clear outbound voice.

[0096] In step S220 of some embodiments, after the noise reduction process, the clear noise-reduced outbound voice can be encoded using a speech coding algorithm to convert the noise-reduced outbound voice into a series of features that can represent the voice signal. The obtained outbound voice features can include Mel-frequency cepstral coefficients (MFCCs), Mel-frequency spectrum energy features, etc., which can capture key information in the voice signal while reducing the dimension of the data.

[0097] In step S230 of some embodiments, further, a speech separation technology can be implemented based on the difference in speech features of different objects to distinguish the features of different speaking objects. In a multi-speaker environment, this can be achieved by using machine learning techniques, such as clustering or classification algorithms, to perform feature separation, with the aim of extracting the features of each independent speaker from the mixed speech. The outbound object speech features are used to indicate the different speaking objects in the historical outbound speech. For example, in the intelligent outbound call scenario of financial technology, the historical outbound speech may include the speech of the intelligent customer service and the speech of the target object, then the outbound object speech features may include the speech features of the intelligent customer service in the outbound speech features and the speech features of the target object in the outbound speech features.

[0098] In step S240 of some embodiments, the separated outbound call object voice features can be further feature decoded to obtain at least one voice segment of the intelligent customer service in the historical outbound call voice and at least one voice segment of the intelligent customer service in the historical outbound call voice. Feature decoding here refers to converting the extracted features back into understandable human voice, and can be specifically done with the help of text-to-speech synthesis (TTS) technology or other synthesis methods.

[0099] In step S250 of some embodiments, since the voice scripts of the intelligent customer service in the historical outbound voice are pre-configured, at least one reply object voice segment of the target object can be selected from the outbound object voice segments based on the historical node script text. The reply object voice segment is the voice segment of the target object replying to the historical node script text in the historical outbound conversation.

[0100] It should be noted that selecting at least one reply object voice segment of the target object from the outbound call object voice segment based on the historical node speech text can specifically include: performing voice-to-text conversion on the outbound call object voice segment to obtain the outbound call object voice segment text; performing text matching on the outbound call object voice segment text and the historical node speech text to obtain the matching status of the outbound call object voice segment text; if the matching status indicates that the outbound call object voice segment text matches the historical node speech text, it is determined that the corresponding outbound call object voice segment belongs to the intelligent customer service, that is, the non-target object; if the matching status indicates that the outbound call object voice segment text does not match the historical node speech text, it is determined that the corresponding outbound call object voice segment belongs to the non-target object. Moreover, this process can rely on content matching and semantic analysis to ensure that the selected segment matches the target object. In this way, all voice segments in which the target object replies to the historical node speech text in the historical outbound call dialogue can be determined.

[0101] In step S260 of some embodiments, further, at least one obtained reply object voice segment can be integrated into voice segments to obtain the historical object voice generated by the target object in the historical outbound call conversation. For example, the historical outbound call voice is "Excuse me, are you XXX? Well, yes. Your insurance policy expires on XX month XX day. Now remind you to complete the payment before the due date to avoid the policy from becoming invalid. OK, got it." After steps S210 to S260 of this application, it can be determined that the reply object voice segment includes "Well, yes" and "OK, got it." After further integration of the voice segments, a complete historical object voice "Well, yes / OK, got it" is formed, wherein " / " is a mark for disconnecting the reply voice of different historical node speech texts.

[0102] It should be noted that the voice segment integration process may adopt Overlap-Add, weighted superposition, etc., to ensure that the reply content of the historical outbound call dialogue is finally perfectly presented.

[0103] In the above embodiment, because the historical outbound voice obtained from a call interaction may contain confirmation results of multiple nodes, it is necessary to first perform voice extraction on the historical outbound voice, so as to more quickly and accurately identify the intention of the target object.

[0104] In step S140 of some embodiments, after obtaining the historical object voice, each reply object voice segment in the historical object voice can be subjected to intention recognition based on the intention recognition model to obtain the node reply intention category and category matching data of the historical task node. Among them, since a historical task node may also contain multiple speech texts, that is, the confirmation of the target object to the historical task node is determined by continuous questioning. The node reply intention category is used to indicate the degree of confirmation of the target object to the historical task node, and the node reply intention category may include a confirmation category, an unconfirmed category, an uncertain category, etc. The confirmation category is that the target object clearly expresses an affirmative intention (such as good, yes, etc.) or gives clear inquiry data (such as giving a clear certificate number, address information, etc.). The unconfirmed category is that the target object clearly expresses a negative intention (such as no, don't know, etc.). The uncertain category is that the target object does not clearly express an affirmative or negative intention (such as unclear, it seems, etc.).

[0105] The category matching data is used to indicate the distance between the node reply intention category and the speech text of the historical object speech. Among them, the category matching data can be directly determined based on the text similarity value between the node reply intention category and the speech text of the historical object speech, and this text similarity value can be calculated based on the Euclidean distance function, the cosine similarity function, etc. The category matching data can also be the confidence (such as the value range [0,1]) obtained after normalizing the obtained text similarity value, and the higher the confidence value, the higher the credibility of the node reply intention category.

[0106] See also Figure 3 , Figure 3 It is a specific flow chart of step S140 provided in an embodiment of the present application. In some embodiments of the present application, step S140 may specifically include but is not limited to steps S310 to S320.

[0107] Step S310, obtaining the node verification category of the historical task node corresponding to the voice segment of the reply object;

[0108] Step S320, perform intent recognition on the reply object voice segment based on the intent recognition model and the node verification category to obtain the node reply intent category and category matching data.

[0109] In step S310 of some embodiments, the node verification category is used to indicate the way to verify whether the historical task node has been confirmed. The node verification category includes an intention recognition category and a labeling category, wherein the intention recognition category indicates that the corresponding historical task node needs to be determined by performing intention recognition on the corresponding reply object voice segment. The labeling category indicates that the corresponding historical task node is determined by manual labeling. For example, when the historical scenario task is a renewal payment reminder task in the insurance field, and the historical task nodes it contains include identity authentication, payment reminder, and whether to transfer to manual, the node verification category corresponding to the identity authentication can be set as the intention recognition category, the node verification category corresponding to the payment reminder verification can be set as the intention recognition category, and the node verification category corresponding to whether to transfer to manual can be set as the intention recognition category. For another example, when the historical task nodes contained in the historical scenario task can include identity authentication, payment reminder, and appointment for the next return visit, the node verification category corresponding to the identity authentication can be set as the intention recognition category, the node verification category corresponding to the payment reminder verification can be set as the intention recognition category, and the node verification category corresponding to the appointment for the next return visit can be set as the labeling category.

[0110] In step S320 of some embodiments, further, the intent recognition of the reply object voice segment can be performed based on the intent recognition model and the node verification category to obtain the node reply intent category and category matching data. In other words, the present application will consider the node verification category of the historical task node to determine what method is used to recognize the corresponding reply object voice segment to determine the corresponding node reply intent category and category matching data.

[0111] It should be noted that, considering that the nodes that need to be confirmed in each scenario are a clear and limited set of intents, for example, the "historical task nodes | node verification categories" contained in historical scenario task 1 may include: {identity confirmation | intent recognition, payment reminder | intent recognition}, therefore, the intent recognition model of this application can adopt a model built based on the FastText model structure to achieve rapid intent recognition. In addition, the intent recognition model can also adopt other model structures, such as convolutional neural network models, etc., without specific limitation.

[0112] In some specific embodiments, the process of performing intent recognition on the reply object voice segment based on the intent recognition model and the node verification category to obtain the node reply intent category and category matching data may specifically include the following two cases:

[0113] If the node verification category is the intent recognition category, the intent recognition is performed on the reply object's voice segment based on the intent recognition model to obtain the node reply intent category and category matching data;

[0114] If the node verification category is the marked category, the preset category is determined to be the node reply intention category of the reply object voice segment, and the preset matching data is determined to be the category matching data of the reply object voice segment.

[0115] In the above embodiment, if the node verification category is an intention recognition category, the intention recognition of the reply object voice segment can be performed based on the intention recognition model to obtain the node reply intention category and category matching data. If the node verification category is a marked category, that is, a manual marking method is adopted, then the preset category can be directly determined as the node reply intention category of the reply object voice segment, and the preset matching data can be determined as the category matching data of the reply object voice segment. For example, if the historical task node is an appointment for the next return visit, the preset category can be a telephone confirmation category, and the category matching data is 0, which means that the target object has not yet completed the confirmation. In other words, if the node verification category is a marked category, it is not in the recognition process of the target scenario task.

[0116] It should be noted that the present application can configure one or more node verification categories for a node, and when multiple node verification categories are configured, verification can be performed based on the priority of the node verification categories.

[0117] In step S150 of some embodiments, further, the confirmation status of the target object for the historical task node can be determined based on the node reply intention category and the category matching data. For example, when the node reply intention category is the confirmation category, the category matching data is the confidence, and the higher the confidence corresponding to the category, the higher the corresponding category credibility. At this time, a confidence threshold (such as 0.6, 0.7, etc.) can be set, and the confidence of different node reply intention categories can be compared with the confidence threshold. The intention recognition results (i.e., the corresponding intention categories) below the confidence threshold are removed as untrustworthy intentions, and only the intention categories above the confidence threshold are retained, and the node confirmation status of the target object is determined based on the intention category corresponding to the maximum confidence above the confidence threshold. Among them, if the intention category corresponding to the maximum confidence is the confirmation category, the node confirmation status is the confirmed state, that is, it is considered that the customer has completed the confirmation of this node; if the intention category corresponding to the maximum confidence is the unconfirmed category, the node confirmation status is the unconfirmed state, that is, it is considered that the customer has not completed the confirmation of this node. For example, the historical task node includes an identity confirmation node. The target object's reply voice in the historical node speech text "Excuse me, are you XXX?" is "Well, yes". Through recognition, the confidence level of the node reply intention category is 0.8, and the confidence threshold is 0.6, then it can be determined that the node confirmation status of the identity confirmation node is confirmed.

[0118] In step S160 of some embodiments, further, the present application can generate a target outbound call script based on the node confirmation status and the scenario task node. The target scenario task is a task after the historical scenario task, that is, the present application can adjust the subsequent script based on the target object's response to the historical outbound call to optimize the opening script of the next outbound call. For example, if the node confirmation status corresponding to the identity confirmation node is confirmed, and the node confirmation status corresponding to the payment reminder node is confirmed, then the next opening script can first say "Hello XXX, I am XX who contacted you last time. I confirmed your identity information as XXXXX by phone last time, and also informed you that the policy expiration date is XXXX. Do you remember?" In this way, the confirmed information can be expressed first in the opening script to avoid the target object from replying to repeated questions multiple times, generate more accurate outbound call scripts, and thus improve the information reach rate of the outbound call task.

[0119] See also Figure 4 , Figure 4 It is a specific flow chart of step S160 provided in an embodiment of the present application. In some embodiments of the present application, step S160 may specifically include but is not limited to steps S410 to S450.

[0120] Step S410, dividing the scene task nodes based on the node confirmation status to obtain target confirmed nodes and target unconfirmed nodes;

[0121] Step S420, obtaining node confirmation data of the target object for the target confirmed node;

[0122] Step S430, generating a speech for the target confirmed node and the node confirmation data based on the speech generation model to obtain an outbound call opening speech;

[0123] Step S440, generating a speech for the target unconfirmed node based on the speech generation model to obtain the speech for the outbound unconfirmed node;

[0124] Step S450, combining the outbound call opening words and the outbound call unconfirmed node words to obtain the target outbound call words.

[0125] In step S410 of some embodiments, the target confirmed node refers to a node in the scene task node whose node confirmation status is confirmed. The target unconfirmed node refers to a node in the scene task node whose node confirmation status is unconfirmed.

[0126] See also Figure 5 , Figure 5It is a specific flow chart of step S410 provided in an embodiment of the present application. In some embodiments of the present application, step S410 may specifically include but is not limited to steps S510 to S530.

[0127] Step S510, selecting a historically confirmed node from the historical task nodes based on the node confirmation status;

[0128] Step S520, performing node matching on the historically confirmed nodes and the scene task nodes to obtain a node matching result;

[0129] Step S530, determining target confirmed nodes and target unconfirmed nodes from the scenario task nodes based on the node matching result.

[0130] In some embodiments, in step S510 to step S530, since the historical scenario task and the target scenario task may be the same or different, the historical task node and the scenario task node may also be the same or different. In this way, the historical confirmed node (i.e., the node whose corresponding node confirmation status is the confirmed state) can be selected from the historical task node based on the node confirmation status, and then the historical confirmed node and the scenario task node are node matched to obtain a node matching result, which is used to indicate whether the scenario task node has a historical task node that has been confirmed to be completed.

[0131] In step S420 of some embodiments, the node confirmation data refers to the relevant object data of the corresponding target confirmed node. For example, if the target confirmed node is an identity confirmation node, then the node confirmation data is the relevant identity information previously stored by the target object, which is used to generate a complete target outbound call script.

[0132] In steps S430 to S450 of some embodiments, further, the speech generation model can be used to combine the confirmed nodes, node confirmation data and target unconfirmed nodes to generate the target outbound call opening speech, that is, the opening speech of the next scenario task. This speech can reflect the information previously confirmed by the target object, making the conversation more fluent and personalized. The speech generation model of the present application can be constructed based on a large model, and prompt text can be generated based on the confirmed nodes, node confirmation data and target unconfirmed nodes, and the prompt text can be input into the speech generation model to obtain the target outbound call speech. For example, the prompt text is "I am AI telephone customer service XX. There is a customer XXX. I need to remind him of [Task 1] [Task 2] and need to confirm [Node 1] [Node 2]... with him. The confirmation of [Node 1] has been completed through the last contact. Please help me redesign the opening remarks of the next outbound call and directly confirm the information of [Node 1] with the customer to avoid asking the customer to answer the same question multiple times." In this way, the generated target outbound call speech can be "Hello Mr. XXX, I am XX who contacted you last time. The last time I called you, I confirmed that your identity information is XXXX. At the same time, I also notified you that the expiration date of the insurance policy is XXXX. Do you still remember it?".

[0133] In the above embodiment, the present application can combine the opening speech, the confirmed node and the unconfirmed node speech to form a complete target outbound call speech. This combined speech will be used in the actual outbound call dialogue, and can flexibly guide the dialogue process according to the different responses of the users, ensuring that the dialogue can effectively cover all key nodes, whether confirmed or unconfirmed, to avoid customers responding to the same questions multiple times, and at the same time, the outbound call speech can be customized for each customer, thereby further improving customer satisfaction.

[0134] See also Figure 6 , Figure 6 It is another specific flow chart of the method for generating speech based on intelligent outbound calls provided by the embodiment of the present application. In some embodiments of the present application, after step S160, the method for generating speech based on intelligent outbound calls of the present application may further include but not limited to steps S610 to S630.

[0135] Step S610, obtaining the task priority of the target scenario task;

[0136] Step S620, matching the target scene tasks with the interaction mode based on the task priority to determine the target interaction mode;

[0137] Step S630, making an intelligent outbound call to the target object based on the target outbound calling speech and the target interaction mode.

[0138] In step S610 of some embodiments, the present application can also intelligently adjust the contact channel for each outbound call, that is, the interaction method. Among them, the interaction method with the target object may include AI outbound calls, APP reminders, manual outbound calls, etc. Task priority is the importance of the current target scenario task that is pre-configured. Task priority can be determined based on multiple factors, such as the urgency and importance of the task, the user's historical response pattern, business rules, etc.

[0139] In step S620 of some embodiments, the target scene task is matched with the target interaction mode based on the task priority, and the target interaction mode of the current target scene task and the target object can be determined. The target interaction mode is used to indicate the communication mode adopted to confirm the target task node to the target object. Among them, the determination of the target interaction mode can depend on various factors, such as the nature of the task, the user's preference, the expected response rate, cost-benefit analysis, etc. For example, for high-priority tasks, the system may choose a more direct and immediate communication mode such as a phone call; while for low-priority tasks, it may choose a lower-cost communication mode such as email or text message.

[0140] See also Figure 7 , Figure 7 It is a specific flow chart of step S620 provided in an embodiment of the present application. In some embodiments of the present application, step S620 may specifically include but is not limited to steps S710 to S740.

[0141] Step S710, obtaining a pre-assigned interaction mode for a target scenario task;

[0142] Step S720, performing saturation division on the preset interaction benchmark number of the pre-assigned interaction mode based on a preset saturation division function to obtain a saturation division interval;

[0143] Step S730, performing interval matching on the number of interaction objects to be allocated of the pre-allocated interaction mode based on the saturation degree sub-interval, and determining a matching interval of the pre-allocated interaction mode;

[0144] Step S740: If the matching interval is the first saturation interval, the pre-assigned interaction mode is adjusted based on the task priority to obtain a target interaction mode.

[0145] In step S710 of some embodiments, considering that there may be an imbalance in the current tasks of different interaction modes, the present application can calculate the saturation of different interaction modes and perform intelligent scheduling according to the saturation, thereby improving the efficiency of intelligent outbound calls. Pre-allocated interaction mode refers to an interaction mode pre-configured for the target scenario task.

[0146] In step S720 of some embodiments, further, the present application may use a preset saturation partitioning function to evaluate the efficiency and capacity of the pre-allocated interactive method. This function can divide different saturation partitioning intervals according to the preset interaction benchmark number M of the pre-allocated interactive method (for example, the average number of insurance policies processed per month, etc.). The saturation level can be understood as the busyness or availability of the interactive method. Among them, based on the preset saturation partitioning function, a first saturation interval and a second saturation interval can be divided, wherein the saturation level of the first saturation interval is higher than the saturation level of the second saturation interval, which means that the interactive method in the first saturation interval is busier or close to its capacity upper limit. For example, the first saturation interval is equivalent to a medium-high saturation interval, and the number of interactive objects to be allocated in the medium-high saturation interval is N≥M*30% (where 30% is the first interval adjustment parameter, which can be flexibly adjusted according to actual needs), the first saturation interval is equivalent to a low saturation interval, and the number of interactive objects to be allocated in the low saturation interval is N <M*30%。

[0147] In step S730 of some embodiments, the number of interaction objects to be allocated refers to data that needs to be communicated using the pre-allocated interaction method but has not yet been allocated, such as the number of insurance policies to be allocated. Further, the number of interaction objects to be allocated of the pre-allocated interaction method can be interval matched based on the saturation sub-interval, and the saturation interval to which the number of interaction objects to be allocated of the pre-allocated interaction method belongs is determined, which is the matching interval of the pre-allocated interaction method.

[0148] In step S740 of some embodiments, if the matching interval is the first saturation interval, that is, the pre-allocated interaction mode of the target scene task belongs to the medium-high saturation interval, then it is currently necessary to adjust the interaction mode of the to-be-allocated interactive object corresponding to the pre-allocated interaction mode based on the task priority of the target scene task to obtain the target interaction mode of the target scene task. Specifically, if the pre-allocated interaction mode of the target scene task belongs to the medium-high saturation interval, and the task priority of the target scene task is greater than the preset priority, then it is necessary to adjust the interaction mode of the pre-allocated interaction mode. If the pre-allocated interaction mode of the target scene task belongs to the medium-high saturation interval, and the task priority of the target scene task is less than or equal to the preset priority, then it is not necessary to adjust the interaction mode of the pre-allocated interaction mode.

[0149] For example, the target scenario task is "renewal payment reminder", the corresponding task priority is 8, the pre-assigned interaction method is "manual outbound call", and the preset priority is 7. According to step S720 and step S730, it is determined that the matching interval of the pre-assigned interaction method of the target scenario task belongs to the first saturation interval, and the task priority of the pre-assigned interaction method is greater than 7. At this time, there is no need to adjust the pre-assigned interaction method of the target scenario task, and "manual outbound call" is used as the target interaction method. If the task priority corresponding to the target scenario task is 7, then it is determined that the matching interval of the pre-assigned interaction method of the target scenario task belongs to the first saturation interval, and the task priority of the pre-assigned interaction method is equal to 7. At this time, the pre-assigned interaction method of the target scenario task needs to be updated and adjusted, and the updated target interaction method can be "AI outbound call".

[0150] Furthermore, the first saturation interval includes a first saturation sub-interval and a second saturation sub-interval, and the saturation degree of the first saturation sub-interval is higher than that of the second saturation sub-interval. At this time, the first saturation sub-interval is equivalent to a high saturation interval, and the second saturation sub-interval is equivalent to a low saturation interval. For example, the first saturation interval is equivalent to a medium-high saturation interval, and the number of interactive objects to be allocated in the medium-high saturation interval is N≥M*30% (where 30% is the first interval adjustment parameter, which can be flexibly adjusted according to actual needs). Further, the second interval adjustment parameter can be 80%. At this time, the number of interactive objects to be allocated in the first saturation sub-interval is N≥M*80% (where 80% is the second interval adjustment parameter, which can be flexibly adjusted according to actual needs), the number of interactive objects to be allocated in the second saturation sub-interval is N≥M*30% and N <M*80%。

[0151] Based on this, step S740 may specifically include the following situations:

[0152] If the matching interval is the first saturated subinterval, the pre-assigned interaction mode is adjusted to obtain a target interaction mode, and the target interaction mode is different from the pre-assigned interaction mode;

[0153] If the matching interval is the second saturated subinterval and the task priority is greater than the preset priority, the pre-assigned interaction mode is used as the target interaction mode;

[0154] If the matching interval is the second saturated subinterval and the task priority is less than or equal to the preset priority, the pre-allocated interaction mode is adjusted to obtain a target interaction mode, and the target interaction mode is different from the pre-allocated interaction mode.

[0155] It should be noted that the present application can divide the saturation partition interval into three intervals based on the preset saturation partition function, namely, the first saturation sub-interval (high saturation interval), the second saturation sub-interval (medium saturation interval) and the second saturation interval (low saturation interval). In addition, the first interval adjustment parameter and the second interval adjustment parameter can also be adjusted to update the number of intervals and the specific division of the intervals, without specific limitation.

[0156] In step S630 of some embodiments, further, after determining the target interaction mode, the previously generated target outbound call speech can be combined to communicate with the target object in the target interaction mode. In this way, the intelligent outbound call of the present application can be carried out according to the preset target outbound call speech and communication strategy, and the response of the target object can be detected in real time, and the speech or interaction strategy can be adjusted as needed. For example, if the target interaction mode is "AI outbound call", if the target object does not answer the call, the system can automatically send a text message reminder.

[0157] In the above embodiments, the present application can enable the intelligent outbound calling system to flexibly select the most appropriate interaction method according to the task priority and characteristics of the target object, thereby improving the efficiency and effectiveness of communication and making outbound calling activities more personalized and goal-oriented.

[0158] It should be noted that the non-Company software tools or components appearing in the embodiments of the present application are merely examples and do not represent actual use.

[0159] The embodiment of the present application provides a method for generating speech based on intelligent outbound calls. The present application proposes a human-machine collaborative outbound call strategy based on user behavior analysis in multiple interactive modes. On the one hand, the opening speech for the next outbound call can be dynamically generated based on the node information confirmed by the customer (i.e., the node confirmation status of the target object for the historical task node is determined based on the node reply intention category and category matching data), thereby avoiding customers from replying to the same type of questions multiple times, and at the same time, customizing personalized outbound call speech for each customer to improve customer satisfaction; on the other hand, by dynamically predicting the saturation level of each interactive mode, the interactive mode with the target object can be adjusted in real time, thereby realizing the collaboration between human and AI and improving the information reach and efficiency of outbound call tasks.

[0160] See also Figure 8 The embodiment of the present application further provides a speech generation device based on intelligent outbound calling, which can implement the above-mentioned speech generation method based on intelligent outbound calling, and the device includes:

[0161] A first acquisition module 810 is used to acquire a target scene task of a target object, where the target scene task includes a scene task node;

[0162] The second acquisition module 820 is used to acquire the historical outbound voice of the target object in the historical scenario task, and the historical scenario task includes the historical task node and the historical node speech text of the historical task node;

[0163] The extraction module 830 is used to perform speech extraction on the historical outbound call speech and the historical node speech text based on a preset speech extraction model to obtain the historical object speech, where the historical object speech is the speech of the target object in replying to the historical node speech text in the historical outbound call dialogue;

[0164] The recognition module 840 is used to perform intent recognition on the historical object speech based on the intent recognition model, and obtain the node reply intent category and category matching data of the historical task node, where the category matching data is used to indicate the distance between the node reply intent category and the speech text of the historical object speech;

[0165] A determination module 850, for determining a node confirmation status of a target object for a historical task node based on the node reply intention category and the category matching data;

[0166] The generation module 860 is used to generate target outbound call speech based on the node confirmation status and the scenario task node.

[0167] The specific implementation of the speech generation device based on intelligent outbound calls in the embodiment of the present application is basically the same as the specific implementation of the speech generation method based on intelligent outbound calls mentioned above, and will not be repeated here.

[0168] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned method for generating speech based on intelligent outbound calls when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0169] See also Fig. 9 , Fig. 9 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0170] The processor 910 may be implemented by a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0171] The memory 920 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 920 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 920, and the processor 910 calls and executes the method for generating speech based on intelligent outbound calls in the embodiment of this application;

[0172] Input / output interface 930, used to implement information input and output;

[0173] Communication interface 940, used to realize communication interaction between the device and other devices, which can be realized through wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0174] bus 950 , which transmits information between the various components of the device (e.g., processor 910 , memory 920 , input / output interface 930 , and communication interface 940 );

[0175] The processor 910 , the memory 920 , the input / output interface 930 , and the communication interface 940 are connected to each other in communication within the device via a bus 950 .

[0176] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the method for generating speech based on intelligent outbound calls is implemented.

[0177] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0178] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0179] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0180] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0181] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0182] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0183] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0184] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0185] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0186] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0187] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0188] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A method for generating speech based on intelligent outbound calling, characterized in that: The method comprises: Acquire a target scenario task of a target object, wherein the target scenario task includes a scenario task node; Obtain the historical outbound voice of the target object in the historical scenario task, wherein the historical scenario task includes a historical task node and a historical node speech text of the historical task node; Performing voice extraction on the historical outbound call voice and the historical node speech text based on a preset voice extraction model to obtain a historical object voice, where the historical object voice is the voice of the target object in replying to the historical node speech text in the historical outbound call conversation; Performing intent recognition on the historical object speech based on the intent recognition model to obtain the node reply intent category and category matching data of the historical task node, wherein the category matching data is used to indicate the distance between the node reply intent category and the speech text of the historical object speech; Determine the node confirmation status of the target object for the historical task node based on the node reply intention category and the category matching data; Generate target outbound call speech based on the node confirmation status and the scenario task node.

2. The method according to claim 1, characterized in that The generating of the target outbound call speech based on the node confirmation state and the scenario task node includes: Divide the scene task nodes into nodes based on the node confirmation status to obtain target confirmed nodes and target unconfirmed nodes; Obtaining node confirmation data of the target object for the target confirmed node; Based on the speech generation model, speech generation is performed on the target confirmed node and the node confirmation data to obtain the outbound call opening speech; Generate speech for the target unconfirmed node based on the speech generation model to obtain the speech for the outbound unconfirmed node; The target outbound call speech is obtained by combining the speech based on the outbound call opening speech and the outbound call unconfirmed node speech.

3. The method according to claim 2, characterized in that The node division of the scene task node based on the node confirmation status to obtain a target confirmed node and a target unconfirmed node includes: Selecting a historical confirmed node from the historical task nodes based on the node confirmation status; Perform node matching on the historically confirmed nodes and the scenario task nodes to obtain a node matching result; The target confirmed node and the target unconfirmed node are determined from the scenario task nodes based on the node matching result.

4. The method according to claim 1, characterized in that The method further comprises: Obtaining the task priority of the target scenario task; Matching the target scenario tasks with interaction modes based on the task priorities to determine a target interaction mode, where the target interaction mode is used to indicate a communication mode used to confirm the target task node to the target object; Based on the target outbound calling speech and the target interaction mode, an intelligent outbound call is made to the target object.

5. The method according to claim 4, characterized in that The step of matching the target scene tasks with the interaction mode based on the task priority and determining the target interaction mode includes: Obtaining a pre-assigned interaction mode for the target scenario task; Performing saturation division on the preset interaction benchmark number of the pre-allocated interaction mode based on a preset saturation division function to obtain saturation division intervals, wherein the saturation division intervals include a first saturation interval and a second saturation interval, and the saturation degree of the first saturation interval is higher than the saturation degree of the second saturation interval; Performing interval matching on the number of interaction objects to be allocated of the pre-allocated interaction mode based on the saturation degree sub-interval, and determining a matching interval of the pre-allocated interaction mode; If the matching interval is the first saturation interval, the pre-allocated interaction mode is adjusted based on the task priority to obtain the target interaction mode.

6. The method according to any one of claims 1 to 5, characterized in that: The method of performing voice extraction on the historical outbound voice and the historical node speech text based on a preset voice extraction model to obtain the historical object voice includes: Performing noise reduction processing on the historical outbound call voice to obtain noise-reduced outbound call voice; Performing speech encoding on the noise-reduced outbound voice to obtain outbound voice features; Performing voice feature separation on the outbound voice feature to obtain voice features of an outbound call recipient, wherein the voice features of the outbound call recipient are used to indicate different speaking recipients in the historical outbound voice; Decoding the voice features of the outbound call recipient to obtain a voice segment of the outbound call recipient; Selecting a reply object voice segment from the outbound call object voice segment based on the historical node speech text, wherein the reply object voice segment is a voice segment in which the target object replies to the historical node speech text in the historical outbound call conversation; The reply object voice segments are integrated to obtain the historical object voice.

7. The method according to claim 6, characterized in that The performing intention recognition on the historical object speech based on the intention recognition model to obtain the node reply intention category and category matching data of the historical task node includes: Obtaining a node verification category of the historical task node corresponding to the reply object voice segment, wherein the node verification category includes an intent recognition category and a labeling category; If the node verification category is the intention recognition category, perform intention recognition on the reply object voice segment based on the intention recognition model to obtain the node reply intention category and the category matching data; If the node verification category is the marked category, the preset category is determined to be the node reply intention category of the reply object voice segment, and the preset matching data is determined to be the category matching data of the reply object voice segment.

8. A speech generation device based on intelligent outbound calling, characterized in that: The device comprises: A first acquisition module, used to acquire a target scene task of a target object, wherein the target scene task includes a scene task node; The second acquisition module is used to acquire the historical outbound voice of the target object in the historical scenario task, wherein the historical scenario task includes a historical task node and a historical node speech text of the historical task node; An extraction module, for performing speech extraction on the historical outbound call speech and the historical node speech text based on a preset speech extraction model to obtain a historical object speech, wherein the historical object speech is the speech of the target object in replying to the historical node speech text in the historical outbound call dialogue; A recognition module, used to perform intent recognition on the historical object speech based on an intent recognition model, and obtain a node reply intent category and category matching data of the historical task node, wherein the category matching data is used to indicate a distance between the node reply intent category and the speech text of the historical object speech; A determination module, configured to determine a node confirmation status of the target object for the historical task node based on the node reply intention category and the category matching data; A generation module is used to generate target outbound call speech based on the node confirmation status and the scenario task node.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images