Conversation processing method and device based on artificial intelligence, equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence and natural language processing, can be applied to the field of intelligent medical treatment and financial science and technology, and discloses a dialogue processing method and device based on artificial intelligence, equipment and a medium. According to the current dialogue data, performing dialogue interruption detection by using a concurrent multi-mode interruption detection mechanism based on audio and text to obtain an interruption judgment result used for determining whether a user has an interruption intention or not; if the interruption judgment result is that the user has the interruption intention, current verbal skill output is stopped, and after user dialogue input is completed, user intention recognition is carried out according to the current input dialogue of the user so as to judge whether the user intention is an original intention or a new intention; and outputting response contents corresponding to different intentions according to the user intention recognition result and a preset mapping relationship. According to the invention, dialogue interruption detection can be realized, the problem of poor experience caused by communication interruption after customer interruption is effectively solved, and the user experience is improved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and natural language processing, and in particular to a dialogue processing method, apparatus, device and medium based on artificial intelligence. Background Technology

[0002] With the development of internet technology, artificial intelligence (AI) technology has also emerged. AI-based conversational interaction has been applied in various scenarios. For example, in smart healthcare, AI-based chatbots guide hospital procedures to help patients quickly understand the corresponding medical processes. In the financial sector, AI-based chatbots assist banks with business recommendations, marketing, and debt collection calls, enabling better customer maintenance and rapid business expansion.

[0003] Existing intelligent chatbots often rely on a predefined "turn-and-go" rhythm and pre-stored dialogue scripts to interact with customers. That is, the intelligent chatbot finishes speaking according to the pre-set script and then the customer responds. However, when the customer interrupts, the intelligent chatbot still outputs a whole paragraph according to the original script, creating the illusion that the customer's input is ignored, or leading to awkward situations such as double-talking and information overlap, resulting in a very poor customer experience. Summary of the Invention

[0004] This invention provides a dialogue processing method, apparatus, device, and medium based on artificial intelligence to solve the technical problem that when a customer interrupts, the intelligent dialogue robot continues to output the original script, causing the customer's input to be ignored or leading to awkward situations such as double-talking and information overlap, resulting in a very poor customer experience.

[0005] Firstly, an artificial intelligence-based dialogue processing method is provided, including:

[0006] Get the current conversation data;

[0007] Based on the current dialogue data, a concurrent multimodal interruption detection mechanism based on audio and text is used to detect dialogue interruptions and obtain an interruption judgment result to determine whether the user has the intention to interrupt.

[0008] If the interruption judgment result indicates that the user has the intention to interrupt, the current speech output is stopped. After the user finishes the dialogue input, the user's intention is identified based on the user's current input dialogue to determine whether the user's intention is the original intention or a new intention.

[0009] Based on the user intent recognition results, the corresponding response content is output according to the pre-set mapping relationship.

[0010] Secondly, an artificial intelligence-based dialogue processing device is provided, comprising:

[0011] The acquisition module is used to acquire the current dialogue data;

[0012] The interruption detection module is used to detect dialogue interruptions based on the current dialogue data using a concurrent multimodal interruption detection mechanism based on audio and text, and to obtain an interruption judgment result to determine whether the user has the intention to interrupt.

[0013] The intent determination module is used to stop the current dialogue output when the interruption judgment result indicates that the user has the intention to interrupt. After the user finishes the dialogue input, the module performs user intent recognition based on the user's current input dialogue to determine whether the user intent is the original intent or a new intent.

[0014] The output processing module is used to output the response content corresponding to different intentions according to the user intent recognition results and the pre-set mapping relationship.

[0015] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described dialogue processing method.

[0016] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the above-described dialogue processing method.

[0017] In the aforementioned AI-based dialogue processing method, apparatus, device, and medium, the current dialogue data can be obtained through the client. Based on this data, a concurrent multimodal interruption detection mechanism using audio and text is employed to detect dialogue interruptions, resulting in an interruption judgment result to determine if the user intends to interrupt. If the interruption judgment result indicates the user intends to interrupt, the current dialogue output is stopped, and the user's dialogue input is completed. User intent is then identified based on the current input dialogue to determine whether the user intent is the original intent or a new intent. Finally, based on the user intent identification result, response content corresponding to different intents is output according to a pre-set mapping relationship, and the response content is fed back to the client. In this invention, for online medical services in smart healthcare and online services in financial and insurance business scenarios, when there is... When dialogues occur, such as consultations and product marketing promotions, a concurrent multimodal interruption detection mechanism based on audio and text can be used to detect interruptions. When a user's intention to interrupt is detected, the current dialogue output is stopped, and the user is allowed to input their dialogue. After input, the user's intent is identified based on the current input dialogue, and then the corresponding response content is output according to the pre-set mapping relationship based on the user intent identification result. This effectively reduces awkwardness, misunderstandings, and information conflicts in the call when interruptions are detected, avoids dialogue misalignment caused by AI forcibly outputting, enhances the human-like performance of the intelligent robot during dialogue, and significantly improves customer goodwill and interaction fluency. This effectively solves the problem of extremely poor experience caused by communication interruption after a customer interrupts, thereby achieving truly natural two-way interaction. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of an application environment for a dialogue processing method according to an embodiment of the present invention;

[0020] Figure 2 This is a flowchart illustrating the dialogue processing method in the first embodiment of the present invention;

[0021] Figure 3 yes Figure 2 A schematic diagram of a specific implementation method for step S102;

[0022] Figure 4 yes Figure 3 A schematic diagram of a sub-process of a specific implementation of step S1021;

[0023] Figure 5 yes Figure 3 Another sub-process diagram of a specific implementation of step S1021;

[0024] Figure 6 yes Figure 3 Another sub-process diagram of a specific implementation of step S1021;

[0025] Figure 7 yes Figure 2 A schematic diagram of a specific implementation of step S104;

[0026] Figure 8 This is a flowchart illustrating the dialogue processing method in the second embodiment of the present invention;

[0027] Figure 9 This is a schematic diagram of a dialogue processing device according to an embodiment of the present invention;

[0028] Figure 10 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention;

[0029] Figure 11 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present 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 the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] The AI-based dialogue processing method provided in this invention can be applied to, for example... Figure 1In this application environment, the client communicates with the server via a network. The server can obtain the current dialogue data from the client and perform dialogue interruption detection using a concurrent multimodal interruption detection mechanism based on audio and text. This results in an interruption judgment result to determine if the user intends to interrupt. If the interruption judgment result indicates that the user intends to interrupt, the current dialogue output is stopped, and the user's dialogue input is completed. Then, user intent recognition is performed based on the user's current input dialogue to determine whether the user intent is the original intent or a new intent. Finally, based on the user intent recognition result, the corresponding response content is output according to a pre-set mapping relationship, and the response content is fed back to the client. In this invention, for online medical services in smart healthcare and online services in business scenarios such as finance and insurance, when there are dialogues such as consultation and product marketing promotion... When interruptions occur, a concurrent multimodal interruption detection mechanism based on audio and text can be used to detect interruptions based on current dialogue data. Upon detecting a user's intention to interrupt, the current dialogue output is stopped, and the system waits for user input. After input, the system identifies the user's intent based on the current dialogue input and outputs responses corresponding to different intents according to a pre-set mapping relationship. This effectively reduces awkwardness, misunderstandings, and information conflicts during calls, avoids dialogue misalignment caused by forced AI output, and enhances the human-like performance of the intelligent robot during dialogue. This significantly improves customer satisfaction and interaction fluency, effectively solving the problem of poor experience caused by communication disruptions after customer interruptions, thus achieving truly natural two-way interaction. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0032] Please see Figure 2 As shown, Figure 2 A flowchart illustrating the AI-based dialogue processing method provided in the first embodiment of the present invention includes the following steps:

[0033] S101: Get the current dialogue data.

[0034] In this step, the current dialogue data is acquired in real time. The dialogue data can be voice call data. If there is no interruption, the voice dialogue is output according to the predetermined script.

[0035] In this embodiment, ASR (Automatic Speech Recognition) technology can be used to acquire current dialogue data in real time. In this embodiment, ASR technology can convert speech into text. Of course, other methods that can achieve the above functions can also be used instead, which will not be elaborated here.

[0036] The AI-based dialogue processing method provided by this invention can be applied to intelligent chatbots in various application scenarios. Intelligent chatbots are typically implemented through a server that can acquire current dialogue data. For example, in online medical consultation scenarios in the smart healthcare field, intelligent chatbots can guide users through registration, obtaining appointment numbers, and evaluating medical services. In product promotion or business marketing scenarios in the fintech field, intelligent chatbots are often used to market products to users, thereby improving the efficiency of product or business promotion.

[0037] S102: Based on the current dialogue data, a concurrent multimodal interruption detection mechanism based on audio and text is used to detect dialogue interruptions and obtain an interruption judgment result to determine whether the user has the intention to interrupt.

[0038] In this step, a concurrent multimodal interruption detection mechanism based on audio and text is used to comprehensively identify and detect whether there is customer interruption behavior in the current dialogue data.

[0039] Specifically, such as Figure 3 As shown, step S102 includes the following steps S1021-S1022:

[0040] S1021. Based on the audio dimension, text dimension and rhythm dimension, the current dialogue data is subjected to dialogue interruption detection in parallel to obtain the interruption confidence scores of the three dimensions.

[0041] In this step, dialogue interruption detection is performed on the current dialogue data in parallel based on three modal dimensions: audio dimension, text dimension, and rhythm dimension.

[0042] like Figure 4 As shown, in this embodiment, step S1021, which involves detecting dialogue interruptions based on the audio dimension of the current dialogue data, includes the following steps S211a-S214a:

[0043] S211a. Perform voiceprint separation on the current dialogue data to obtain the user's voice signal and the AI ​​voice signal.

[0044] In this step, pre-trained deep learning models for speaker speech separation, such as TasNet (Time-Domain Audio Network), DPRNN (Dual-Path Recurrent Neural Network), SepFormer, etc., can be used to perform speaker separation on the current dialogue data, separating the speech streams of different speakers from the mixed audio to obtain the user speech signal and the AI ​​speech signal.

[0045] S212a. Overlap detection is performed on the user's voice signal and the AI ​​voice signal to obtain the time axis crossover and spectral overlap.

[0046] In this step, endpoint detection (VAD) is performed on the separated user voice signal and AI voice signal. The voice activity time periods are compared to determine whether there is overlap (time axis intersection). Then, the spectral similarity (i.e. spectral overlap) is calculated in the overlapping time periods.

[0047] S213a. Monitor the loudness of the user's voice when the AI ​​voice is output based on the prosodic features of the user's voice signal and the AI ​​voice signal, and obtain the loudness of the user's voice.

[0048] In this step, loudness features are extracted from the separated user speech signal within the time axis of the AI ​​speech signal. The loudness features can be short-term energy.

[0049] S214a. Obtain the interruption confidence score of the audio dimension based on the time axis crossover, spectral overlap, and user voice loudness.

[0050] For steps S211a-S214a, time axis crossover is a necessary condition for dialogue interruption. The higher the loudness of the user's voice, the higher the interruption confidence score. However, the higher the spectral similarity, the more likely there is incomplete voiceprint separation, which affects the interruption confidence score in the audio dimension, showing a negative correlation. The interruption confidence score is obtained by combining time axis crossover, spectral overlap, and user voice loudness in the above steps. Understandably, machine learning models, such as SVM or Random Forest, can be used to output the interruption confidence score.

[0051] Reference Figure 5 ,like Figure 5 As shown, in this embodiment, step S1021, the dialogue interruption detection based on the text dimension of the current dialogue data includes the following steps S211b-S212b:

[0052] S211b: Convert the current dialogue data into text information.

[0053] S212b. The text information is subjected to interruption keyword identification, and the dialogue interruption is detected based on the frequency of the identified interruption keywords to obtain the interruption confidence score in the text dimension.

[0054] In this embodiment, the interruption keywords may include keywords such as "but", "however", "wait a minute", "I disagree", "I want to ask...", "don't say it yet..." and "let me say it". When the frequency of the interruption keywords identified within a preset time exceeds a preset number threshold, it can be said that the user has the intention to interrupt, and the interruption confidence score is greater than a preset score threshold (such as 0.8).

[0055] Reference Figure 6 ,like Figure 6 As shown, in this embodiment, step S1021, the dialogue interruption detection based on the text dimension of the current dialogue data includes the following steps S211c-S213c:

[0056] S211c: Perform voiceprint separation on the current dialogue data to obtain the user's voice signal and the AI ​​voice signal.

[0057] S212c: Compare the time axis of the user's voice signal and the time axis of the AI ​​voice signal, and calculate the insertion frequency of the user's voice on the time axis of the AI ​​voice signal.

[0058] In this step, endpoint detection (VAD) is performed on the separated user voice signal and AI voice signal to obtain the timeline of their respective voice activities. Then, the user voice segments are traversed to detect whether they fall within the timeline interval of the AI ​​voice (e.g., when the user voice starts, the AI ​​is speaking). If a user voice segment falls within the timeline interval of the AI ​​voice, it is counted as a "user voice insertion event". The number of "user voice insertion events" is counted to obtain the insertion frequency.

[0059] S213c: Obtain the interruption confidence score of the rhythm dimension based on the insertion frequency within a preset time period.

[0060] S1022. Determine whether the user intends to interrupt based on the interruption confidence scores of the three dimensions, and obtain the interruption judgment result.

[0061] In this invention, a parallel detection mechanism is adopted, that is, simultaneous listening. It can comprehensively judge whether the user has the intention to interrupt based on the interruption confidence score of three dimensions, or independently judge whether the user has the intention to interrupt based on the interruption confidence score of each dimension. For example, in the online medical service evaluation scenario in the field of smart healthcare, when the intelligent chatbot asks the user for a rating or experience of the online medical process, if the interruption confidence score of any one of the audio, text and rhythm dimensions is greater than a preset threshold (e.g., 0.8), or if the interruption confidence score of any two of the three dimensions exceeds a medium preset threshold (e.g., 0.6), it indicates that a user interruption event has occurred, the user's intention to interrupt has been detected, and the output of the script needs to be stopped.

[0062] S103: If the interruption judgment result indicates that the user has the intention to interrupt, stop the current dialogue output, wait for the user to finish the dialogue input, and perform user intent recognition based on the user's current input dialogue to determine whether the user intent is the original intent or a new intent.

[0063] In this step, when the user intends to interrupt, the current dialogue output is stopped. Furthermore, a connecting dialogue can be generated based on the contextual information of the current dialogue data. For example, concise, non-interrupting connecting dialogues such as "Okay, you go first" or "Of course, I'll listen to your concerns" can be generated. The dialogue is then waited for by the user, maintaining a "listening and responding" communication rhythm. After the user has finished inputting, the user's intent can be identified based on the current input to obtain the user intent identification result.

[0064] S104: Based on the user intent recognition result, output the response content corresponding to different intents according to the pre-set mapping relationship.

[0065] Specifically, such as Figure 7 As shown, step S104 in this embodiment includes the following steps S1041-S1043:

[0066] S1041. If the user's intent is a new intent, retrieve the matching dialogue according to the pre-set mapping relationship and output the retrieved matching dialogue in response.

[0067] In this step, if the user's intent is a new intent, the corresponding matching script can be retrieved from the knowledge base or script database according to the pre-set mapping relationship.

[0068] In this invention, new intentions can include categories such as asking questions, expressing objections (e.g., expressing objections to the medical service process when evaluating medical services), and expressing concerns (e.g., concerns about product prices). Understandably, new intentions can be dialogue intentions with a topic different from the original dialogue topic. For example, in a dialogue where an intelligent chatbot introduces the features of a medical product, if the user expresses concerns about the product's price, the dialogue topic needs to change from introducing product features to introducing product prices. Alternatively, if the user asks if there are other types of medical products to introduce, the subject involved in the dialogue topic changes, and in this case, the corresponding response needs to be retrieved and output to continue the dialogue.

[0069] S1042. After the new intention dialogue response is completed, the original dialogue text is called according to the context information of the speech with the interruption intention and the pre-set mapping relationship to continue the original intention dialogue.

[0070] In this embodiment, after the new intent response is completed, if there is no new interruption event, the conversation returns to the original topic. Preferably, when returning to the original topic, the context manager can be used to generate connecting words, which are then concatenated with the original topic words for output. For example, "Let me continue talking about the features of this product that I just mentioned to you. I just mentioned that it is..." or "We were halfway through talking, so let's go back to the discount part...".

[0071] S1043. If the user's intent is the original intent, generate a connecting statement based on the context information of the current dialogue data, and call the original statement text according to the pre-set mapping relationship. Then, concatenate the connecting statement text and the original statement text and output the result.

[0072] In this embodiment, the original intent refers to the dialogue intent that is the same as the original dialogue topic. For example, in a medical product recommendation dialogue, the intelligent chatbot introduces the features of the medical product, and the user inquires about the specific features. When the user interrupts the dialogue and pauses briefly without inputting a new intent or changing the original topic, the context manager can be used to generate natural transitional phrases to continue the original topic dialogue output.

[0073] For steps S1041-S1043, the corresponding response content is output according to the user intent recognition result and the pre-set mapping relationship, so as to more accurately locate customer pain points, proactively guide new topics, and the output of the corresponding response content can also be combined with connecting words, which can enhance customers' trust in AI sales and natural acceptance, thereby improving the efficiency of demand discovery and conversion.

[0074] As can be seen, in the above solution, for online medical services in smart healthcare and online services in business scenarios such as finance and insurance, when dialogue behaviors such as consultation and product marketing promotion occur, a concurrent multimodal interruption detection mechanism based on audio and text can be used to detect dialogue interruptions based on the current dialogue data. When the user's intention to interrupt is detected, the current dialogue output is stopped, and the dialogue input is waited for. After the input is completed, the user's intention is identified based on the current input dialogue, and then the response content corresponding to different intentions is output according to the pre-set mapping relationship based on the user's intention identification result. In this way, when the user interruption is detected, it can effectively reduce embarrassment, misunderstanding and information conflict in the call, avoid dialogue misalignment caused by AI forcibly outputting, enhance the human-like performance of the intelligent robot during the dialogue, and significantly improve customer goodwill and interaction fluency. This effectively solves the problem of extremely poor experience caused by communication interruption after the customer interrupts, thereby achieving truly natural two-way interaction.

[0075] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0076] Reference Figure 8 , Figure 8 A flowchart illustrating an artificial intelligence-based dialogue processing method provided in the second embodiment of the present invention includes the following steps:

[0077] S201. Obtain the current dialogue data.

[0078] This step is the same as or similar to step S101, and will not be repeated here.

[0079] S202. Based on the current dialogue data, a concurrent multimodal interruption detection mechanism based on audio and text is used to detect dialogue interruptions and obtain an interruption judgment result to determine whether the user has the intention to interrupt.

[0080] This step is the same as or similar to step S102, and will not be described again here.

[0081] S203. If the interruption judgment result indicates that the user has the intention to interrupt, stop the current dialogue output, wait for the user to finish the dialogue input, and perform user intent recognition based on the user's current input dialogue to determine whether the user intent is the original intent or a new intent.

[0082] This step is the same as or similar to step S103, and will not be repeated here.

[0083] S204. Based on the user intent recognition result, output the response content corresponding to different intents according to the pre-set mapping relationship.

[0084] This step is the same as or similar to step S104, and will not be repeated here.

[0085] S205. Mark the interruption points in the dialogue data where there is an intention to interrupt, and obtain the number of interruptions in this dialogue.

[0086] In this step, user interruption events are marked and recorded, and the number of interruptions in this conversation is recorded. The number of interruptions can be used to help set personalized tags for users.

[0087] S206. Semantic extraction and context aggregation are performed on the utterances in this dialogue data that contain the intention to interrupt, forming a structured semantic summary that includes the category and topic of the interruption intention.

[0088] In this embodiment, the interruption intent categories include new intent and original intent. New intent may include categories such as asking questions, objecting, and expressing concerns.

[0089] In this step, large models or rule engines can be used for semantic extraction and context aggregation to form a structured semantic summary, which can be used to optimize subsequent dialogue strategies.

[0090] S207. Generate a customer profile based on the number of interruptions and the structured semantic summary.

[0091] In this invention, customer profiles can be introduced as personalized tags in subsequent dialogue interactions, allowing for adjustments to dialogue strategies to suit different customer groups. For example, the length of the dialogue can be adjusted, the style of the language can be changed, and price negotiations can be postponed. For instance, if there are many interruptions, the customer profile can include a "prone to interruption" tag, and the subsequent dialogue interaction strategy needs to control the pace of the language and prevent lengthy output. If the user's intention category when interrupting multiple times is a new intention expression concern category, the customer profile can include a "high-risk concern" tag, and the language in the subsequent dialogue interaction strategy needs to focus on risk mitigation and establishing a sense of security.

[0092] As can be seen, in the above solution, for online medical services in smart healthcare and online services in business scenarios such as finance and insurance, when dialogue behaviors such as consultation and product marketing promotion occur, the concurrent multimodal interruption detection mechanism can also be used to detect dialogue interruptions. When a user interruption is detected, the response content corresponding to different intentions can be output according to the pre-set mapping relationship, so as to effectively reduce embarrassment, misunderstanding and information conflict in the call, avoid dialogue misalignment caused by AI forcibly outputting, enhance the human-like performance of the intelligent robot in the dialogue, significantly improve customer goodwill and interaction fluency, and achieve truly natural two-way interaction.

[0093] In one embodiment, an AI-based dialogue processing device is provided, which corresponds one-to-one with the AI-based dialogue processing method in the second embodiment described above. For example... Figure 9 As shown, the dialogue processing device includes an acquisition module 110, an interruption detection module 120, an intent determination module 130, and an output processing module 140. Detailed descriptions of each functional module are as follows:

[0094] Module 110 is used to acquire current dialogue data;

[0095] Interruption detection module 120 is used to detect dialogue interruptions based on the current dialogue data using a concurrent multimodal interruption detection mechanism based on audio and text, and to obtain an interruption judgment result for determining whether the user has an intention to interrupt.

[0096] The intent determination module 130 is used to stop the current dialogue output when the interruption judgment result is that the user has the intention to interrupt. After the user finishes the dialogue input, the module performs user intent recognition based on the user's current input dialogue to determine whether the user intent is the original intent or a new intent.

[0097] The output processing module 140 is used to output the response content corresponding to different intentions according to the user intention recognition result and the pre-set mapping relationship.

[0098] In one embodiment, the interruption detection module 120 is specifically used for:

[0099] Dialogue interruption detection is performed on the current dialogue data in parallel based on audio, text, and rhythm dimensions, and interruption confidence scores are obtained in three dimensions.

[0100] The interruption confidence score is used to determine whether the user intends to interrupt, and the interruption judgment result is obtained.

[0101] In one embodiment, the interruption detection module 120 is specifically used for:

[0102] Perform voiceprint separation on the current dialogue data to obtain the user's voice signal and the AI's voice signal;

[0103] Overlap detection is performed on user voice signals and AI voice signals to obtain time axis crossover and spectral overlap.

[0104] The loudness of the user's voice is obtained by monitoring the prosodic features of the user's voice signal and the AI ​​voice signal during the AI ​​voice output.

[0105] The interruption confidence score for the audio dimension is obtained based on the time axis crossover, spectral overlap, and user voice loudness.

[0106] In one embodiment, the interruption detection module 120 is specifically used for:

[0107] Convert the current dialogue data into text information;

[0108] The text information is subjected to interruption keyword identification, and dialogue interruption detection is performed based on the frequency of the identified interruption keywords to obtain the interruption confidence score in the text dimension.

[0109] In one embodiment, the interruption detection module 120 is further specifically used for:

[0110] Perform voiceprint separation on the current dialogue data to obtain the user's voice signal and the AI's voice signal;

[0111] Compare the time axis of the user's voice signal and the time axis of the AI ​​voice signal, and calculate the insertion frequency of the user's voice on the time axis of the AI ​​voice signal.

[0112] The interruption confidence score of the rhythm dimension is obtained based on the insertion frequency within a preset time period.

[0113] In one embodiment, the output processing module 140 is specifically used for:

[0114] If the user's intent is a new intent, retrieve the matching script according to the pre-set mapping relationship and output the retrieved matching script in response;

[0115] After the new intent dialogue response is completed, the original dialogue text is called according to the context information of the speech with the interruption intent and the pre-set mapping relationship to continue the original intent dialogue;

[0116] If the user's intent is the original intent, a connecting statement is generated based on the context information of the current dialogue data, and the original statement text is called according to the pre-set mapping relationship. The connecting statement and the original statement text are then concatenated and processed before being output.

[0117] In one embodiment, the output processing module 140 is further configured to:

[0118] Mark the interruption points in the dialogue data where there is an intention to interrupt, and obtain the number of interruptions in this dialogue;

[0119] Semantic extraction and context aggregation were performed on the utterances containing interruption intent in this dialogue data to form a structured semantic summary containing interruption intent categories and topics;

[0120] A customer profile is generated based on the number of interruptions and the structured semantic summary.

[0121] This invention provides an AI-based dialogue processing device. First, it uses a concurrent multimodal interruption detection mechanism based on audio and text to detect interruptions in the current dialogue data, obtaining an interruption judgment result to determine if the user intends to interrupt. When the user intends to interrupt, the current dialogue output is stopped, waiting for user input. After input, the user's intent is identified based on the current input dialogue. Then, according to the user intent identification result, different responses are output according to a pre-set mapping relationship. This effectively reduces awkwardness, misunderstandings, and information conflicts during calls when user interruptions are detected, avoids dialogue misalignment caused by forced AI output, and enhances the human-like performance of the intelligent robot during dialogue. This significantly improves customer satisfaction and interaction fluency, effectively solving the problem of poor experience caused by communication breaks after customer interruptions. The dialogue processing device of this invention achieves a near-human salesperson's "interruption detection—letting the conversation (waiting for user input)—natural transition response" capability, realizing truly natural two-way interaction.

[0122] For specific limitations regarding AI-based dialogue processing devices, please refer to the limitations of AI-based dialogue processing methods above, which will not be repeated here. Each module in the aforementioned dialogue processing device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in a computer device, or stored in software in the memory of a computer device, so that the processor can call and execute the operations corresponding to each module.

[0123] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 10 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a server-side artificial intelligence-based dialogue processing method.

[0124] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 11As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements client-side functions or steps of an artificial intelligence-based dialogue processing method.

[0125] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps:

[0126] Get the current conversation data;

[0127] Based on the current dialogue data, a concurrent multimodal interruption detection mechanism based on audio and text is used to detect dialogue interruptions and obtain an interruption judgment result to determine whether the user has the intention to interrupt.

[0128] If the interruption judgment result indicates that the user has the intention to interrupt, the current speech output is stopped. After the user finishes the dialogue input, the user's intention is identified based on the user's current input dialogue to determine whether the user's intention is the original intention or a new intention.

[0129] Based on the user intent recognition results, the corresponding response content is output according to the pre-set mapping relationship.

[0130] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0131] Get the current conversation data;

[0132] Based on the current dialogue data, a concurrent multimodal interruption detection mechanism based on audio and text is used to detect dialogue interruptions and obtain an interruption judgment result to determine whether the user has the intention to interrupt.

[0133] If the interruption judgment result indicates that the user has the intention to interrupt, the current speech output is stopped. After the user finishes the dialogue input, the user's intention is identified based on the user's current input dialogue to determine whether the user's intention is the original intention or a new intention.

[0134] Based on the user intent recognition results, the corresponding response content is output according to the pre-set mapping relationship.

[0135] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0136] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0137] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0138] The above-described embodiments are merely illustrative of the technical solutions of the present invention and are not intended to limit it. 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. 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, and should all be included within the protection scope of the present invention. Furthermore, any software tools or components not belonging to this company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use.

Claims

1. A dialogue processing method based on artificial intelligence, characterized in that, include: Get the current conversation data; Based on the current dialogue data, a concurrent multimodal interruption detection mechanism based on audio and text is used to detect dialogue interruptions and obtain an interruption judgment result to determine whether the user has the intention to interrupt. If the interruption judgment result indicates that the user has the intention to interrupt, the current speech output is stopped. After the user finishes the dialogue input, the user's intention is identified based on the user's current input dialogue to determine whether the user's intention is the original intention or a new intention. Based on the user intent recognition results, the corresponding response content is output according to the pre-set mapping relationship.

2. The dialogue processing method based on artificial intelligence as described in claim 1, characterized in that, The step of detecting dialogue interruptions using a concurrent multimodal interruption detection mechanism based on audio and text, based on the current dialogue data, to obtain an interruption judgment result for determining whether the user intends to interrupt, includes: Dialogue interruption detection is performed on the current dialogue data in parallel based on audio, text, and rhythm dimensions, and interruption confidence scores are obtained in three dimensions. The interruption confidence score is used to determine whether the user intends to interrupt, and the interruption judgment result is obtained.

3. The dialogue processing method based on artificial intelligence as described in claim 2, characterized in that, The method of detecting dialogue interruptions based on the audio dimension of the current dialogue data includes: Perform voiceprint separation on the current dialogue data to obtain the user's voice signal and the AI's voice signal; Overlap detection is performed on user voice signals and AI voice signals to obtain time axis crossover and spectral overlap. The loudness of the user's voice is obtained by monitoring the prosodic features of the user's voice signal and the AI ​​voice signal during the AI ​​voice output. The interruption confidence score for the audio dimension is obtained based on the time axis crossover, spectral overlap, and user voice loudness.

4. The dialogue processing method based on artificial intelligence as described in claim 2, characterized in that, The method of detecting dialogue interruptions based on the text dimension of the current dialogue data includes: Convert the current dialogue data into text information; The text information is subjected to interruption keyword identification, and dialogue interruption detection is performed based on the frequency of the identified interruption keywords to obtain the interruption confidence score in the text dimension.

5. The AI-based dialogue processing method as described in claim 2, characterized in that, The method of detecting dialogue interruptions based on the rhythm dimension of the current dialogue data includes: Perform voiceprint separation on the current dialogue data to obtain the user's voice signal and the AI's voice signal; Compare the time axis of the user's voice signal and the time axis of the AI ​​voice signal, and calculate the insertion frequency of the user's voice on the time axis of the AI ​​voice signal. The interruption confidence score of the rhythm dimension is obtained based on the insertion frequency within a preset time period.

6. The AI-based dialogue processing method as described in claim 1, characterized in that, The step of outputting response content corresponding to different intents according to the user intent recognition result and a pre-set mapping relationship includes: If the user's intent is a new intent, retrieve the matching script according to the pre-set mapping relationship and output the retrieved matching script in response; After the new intent dialogue response is completed, the original dialogue text is called according to the context information of the speech with the interruption intent and the pre-set mapping relationship to continue the original intent dialogue; If the user's intent is the original intent, a connecting statement is generated based on the context information of the current dialogue data, and the original statement text is called according to the pre-set mapping relationship. The connecting statement and the original statement text are then concatenated and processed before being output.

7. The AI-based dialogue processing method as described in any one of claims 1-6, characterized in that, The AI-based dialogue processing method also includes: Mark the interruption points in the dialogue data where there is an intention to interrupt, and obtain the number of interruptions in this dialogue; Semantic extraction and context aggregation were performed on the utterances containing interruption intent in this dialogue data to form a structured semantic summary containing interruption intent categories and topics; A customer profile is generated based on the number of interruptions and the structured semantic summary.

8. A dialogue processing device based on artificial intelligence, characterized in that, include: The acquisition module is used to acquire the current dialogue data; The interruption detection module is used to detect dialogue interruptions based on the current dialogue data using a concurrent multimodal interruption detection mechanism based on audio and text, and to obtain an interruption judgment result to determine whether the user has the intention to interrupt. The intent determination module is used to stop the current dialogue output when the interruption judgment result indicates that the user has the intention to interrupt. After the user finishes the dialogue input, the module performs user intent recognition based on the current input dialogue to determine whether the user intent is the original intent or a new intent. The output processing module is used to output the response content corresponding to different intentions according to the user intent recognition results and the pre-set mapping relationship.

9. A computer 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 dialogue processing method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the dialogue processing method as described in any one of claims 1 to 7.

Images