A real-time analysis method and system for intelligent business

Abstract

A real-time analysis method and system for intelligent business operations, relating to the field of business data analysis, is disclosed. This method, applied to a server, includes: performing time-domain transformation on customer service call voice data to obtain an emotion intensity waveform; segmenting the emotion intensity waveform to obtain multiple emotion features; assessing the reliability of these multiple emotion features to obtain their confidence levels; classifying the multiple emotion features into multiple high-reliability emotion features and multiple low-reliability emotion features based on a pre-set confidence threshold; matching the text of the multiple high-reliability emotion features with the text of the multiple low-reliability emotion features from a voice-text mapping table; and inputting the text of the multiple high-reliability emotion features into a text analysis module to perform customer satisfaction evaluation, obtaining the satisfaction scores corresponding to the multiple high-reliability emotion features. This addresses the problem of low analysis efficiency in high-concurrency scenarios for intelligent voice business analysis methods.

Description

Technical Field

[0001] This application relates to the technical field of business data analysis, specifically to a real-time analysis method and system for intelligent business. Background Technology

[0002] With the rapid development of intelligent customer service and voice analysis technologies, real-time business analysis methods based on voice data play an important role in improving customer service efficiency, optimizing resource allocation, and supporting decision-making.

[0003] Current intelligent analysis methods for voice services extract text and emotional features from customer service call data, and use context encoders and convolutional neural network models for feature extraction and analysis to determine customer satisfaction with the service, thus providing a reference for higher-quality customer service in the future. However, when processing text semantics and emotional features, this technical solution involves multiple sequential steps, such as speech-to-text conversion, text segmentation and encoding, speech denoising, and waveform feature extraction. For example, emotional feature extraction requires denoising by an autoencoder and then conversion into a low-dimensional vector by a CNN. Each step must wait for the previous step to complete before it can be executed, resulting in a strong input-output dependency. This leads to significant latency accumulation of large amounts of data in high-concurrency scenarios, affecting the overall analysis efficiency. Summary of the Invention

[0004] To address the issue of low analysis efficiency in high-concurrency scenarios of current intelligent analysis methods for voice services, this application provides a real-time analysis method and system for intelligent services.

[0005] Firstly, this application provides a real-time analysis method for intelligent services, applied to a server, the method comprising:

[0006] The customer service call voice data is converted in the time domain to obtain an emotion intensity waveform.

[0007] The emotion intensity fluctuation graph is segmented to obtain multiple emotion features;

[0008] A reliability assessment is performed on multiple emotional features to obtain the confidence levels of the multiple emotional features;

[0009] Based on a preset reliability threshold, the multiple emotional features are divided into multiple high-reliability emotional features and multiple low-reliability emotional features.

[0010] From the speech-text mapping table, match texts with multiple high-reliability emotion features and texts with multiple low-reliability emotion features;

[0011] The text of multiple highly reliable emotional features is input into the text analysis module to evaluate customer satisfaction and obtain the satisfaction scores corresponding to the multiple highly reliable emotional features.

[0012] Optionally, the step of performing time-domain conversion on the customer service call voice data to obtain an emotion intensity waveform is as follows:

[0013] The customer service call voice data is processed into frames to obtain multiple data frame sequences;

[0014] The emotional energy of multiple data frame sequences is calculated and normalized to obtain the emotional intensity fluctuation curve.

[0015] Optionally, before performing time-domain conversion on the customer service call voice data to obtain the emotion intensity waveform, the method further includes:

[0016] Identify multiple fluctuation periodic data sequences of the customer service call voice data, wherein all data points in a fluctuation periodic data sequence belong to the same periodic function;

[0017] Calculate the volatility index of multiple said volatility cycle data sequences;

[0018] Based on the volatility index of multiple volatility cycle data sequences, the optimization strategies corresponding to each of the multiple volatility cycle data sequences are matched from a preset optimization strategy table;

[0019] Based on the optimization strategies corresponding to each of the multiple fluctuation cycle data sequences, the multiple fluctuation cycle data sequences are optimized to obtain optimized customer service call voice data.

[0020] Optionally, the step of segmenting the emotion intensity fluctuation map to obtain multiple emotion features specifically includes:

[0021] The emotional intensity fluctuation map is converted into multiple emotional feature maps, including an energy feature map, a frequency feature map, and a fluctuation feature map.

[0022] The multiple emotion feature maps are segmented to obtain multiple sub-emotion features corresponding to each of the multiple emotion feature maps, wherein each sub-emotion feature corresponds to a time window;

[0023] The multiple sub-emotional features corresponding to each of the multiple emotion feature maps are time-series aligned, and the mean of the time window is fused to obtain the emotion intensity fluctuation map segmentation strategy.

[0024] Based on the emotion intensity fluctuation map segmentation strategy, the emotion intensity fluctuation map is segmented into features to obtain multiple emotion features.

[0025] Optionally, the reliability assessment of multiple emotional features to obtain the confidence levels of multiple emotional features specifically involves:

[0026] Based on multiple preset reliability assessment criteria, the reliability of multiple emotional characteristics is assessed to obtain multiple reliability assessment criteria scores corresponding to each of the multiple emotional characteristics. The multiple reliability assessment criteria include stability, consistency and physiological rationality.

[0027] Based on the reliability weights of the multiple reliability assessment criteria, a weighted average is performed on the scores of the multiple reliability assessment criteria corresponding to each of the multiple emotional features to obtain the confidence level of the multiple emotional features.

[0028] Optionally, the step of dividing the multiple emotional features into multiple high-reliability emotional features and multiple low-reliability emotional features according to a preset reliability threshold further includes:

[0029] Based on the customer service call voice data, a customer emotion expression profile is constructed, which includes the base frequency range, volume range, and speech rate range.

[0030] The confidence adjustment coefficient is obtained by calculating the difference between the customer's emotional expression profile and the preset standard emotional expression profile.

[0031] Based on the confidence adjustment coefficient, the confidence threshold corresponding to the standard emotional expression profile is adjusted to obtain the confidence threshold corresponding to the customer service call voice data.

[0032] Optionally, the step of inputting the text of multiple highly reliable emotional features into a text analysis module to perform customer satisfaction evaluation and obtain satisfaction scores corresponding to the multiple highly reliable emotional features further includes:

[0033] Multiple texts with low reliability emotions are input into the lightweight speech verification model of the speech analysis module for text verification, resulting in multiple texts that pass and multiple texts that fail.

[0034] Multiple texts are input into the text analysis module to evaluate customer satisfaction and obtain satisfaction scores corresponding to the multiple texts.

[0035] The speech segments corresponding to the multiple rejected texts are input into a high-precision speech recognition model to obtain the corrected texts corresponding to each of the multiple rejected texts;

[0036] The corrected texts corresponding to each of the multiple failed texts are input into the text analysis module to perform customer satisfaction evaluation and obtain the satisfaction scores corresponding to the multiple corrected texts.

[0037] Secondly, this application provides a real-time analysis system for intelligent services. The system is a server, and the server includes an acquisition module and an analysis module, wherein:

[0038] The acquisition module is used to perform time-domain conversion on the customer service call voice data to obtain an emotion intensity waveform.

[0039] The analysis module is used to perform feature segmentation on the emotion intensity fluctuation map to obtain multiple emotion features; to perform reliability assessment on the multiple emotion features to obtain the confidence level of the multiple emotion features; to divide the multiple emotion features into multiple high-reliability emotion features and multiple low-reliability emotion features according to a preset confidence threshold; to match the text of the multiple high-reliability emotion features and the text of the multiple low-reliability emotion features from the speech-text mapping table, and input the text of the multiple high-reliability emotion features into the text analysis module to perform customer satisfaction evaluation and obtain the satisfaction score corresponding to the multiple high-reliability emotion features.

[0040] Thirdly, this application provides an electronic device including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of the first aspects.

[0041] Fourthly, this application provides a computer-readable storage medium storing instructions that, when executed, perform the method described in any one of the first aspects.

[0042] In summary, one or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0043] This application converts business call audio into an emotion intensity waveform to reflect the customer's emotional state; higher emotion intensity indicates greater emotional agitation. The emotion waveform is then segmented into multiple emotion features, each reflecting the customer's emotional state at that time. However, due to varying communication environments, background noise can be reflected in the emotion features, causing distortion. Therefore, a reliability assessment is performed on each emotion feature to determine its confidence level. Emotion features with high confidence levels are less affected by background noise and have higher recognition accuracy. In this case, the text corresponding to the high-confidence emotion feature can be directly matched from the speech-text mapping table and input directly into the text analysis module for customer satisfaction evaluation. This process breaks through the traditional strong input-output dependency, enabling parallel operation of speech recognition and text semantic analysis, thus significantly improving the efficiency of real-time business analysis of voice data in high-concurrency scenarios. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating a real-time analysis method for intelligent services provided in an embodiment of this application.

[0045] Figure 2 This is a waveform diagram of customer call voice data provided in an embodiment of this application.

[0046] Figure 3 This application provides a schematic diagram of the structure of a real-time analysis system for intelligent services.

[0047] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0048] Explanation of reference numerals in the attached diagram: 1. Acquisition module; 2. Analysis module; 400. Electronic device; 401. Processor; 402. Communication bus; 403. User interface; 404. Network interface; 405. Memory. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0050] Current intelligent analysis methods for voice services still follow a fixed, sequential process. The core steps mainly include voice analysis and text analysis. Voice analysis converts speech into text, while text analysis extracts keywords from the converted text and performs targeted analysis (satisfaction analysis, demand analysis, etc.) on the extracted keywords. To cope with high-concurrency scenarios, most methods allocate more analysis lines for processing. For example, when a data congestion occurs on one analysis line, another new analysis line is allocated to distribute the congested data and alleviate the processing pressure on a single line. However, this approach still does not escape the strong dependency between input and output. When the data volume increases significantly and reaches the upper limit of the processing capacity of all analysis lines, data latency will still accumulate, resulting in a significant decrease in overall analysis efficiency.

[0051] To address the aforementioned problems, this application provides a real-time analysis method for intelligent services. This method is applied to a server, such as... Figure 1 As shown, the method includes steps S101 to S106, which are as follows:

[0052] S101. Perform time-domain conversion on the customer service call voice data to obtain an emotion intensity waveform.

[0053] In the above steps, the voice recording of the customer's conversation with the customer service representative is collected by the voice acquisition module. The voice recording is then converted into digital customer service voice data. The customer's voice data is then converted in the time domain to obtain a two-dimensional emotion intensity waveform with time as the horizontal axis and emotion intensity as the vertical axis.

[0054] In one possible implementation, the emotion intensity waveform essentially reflects the customer's emotional state through the volume and continuity of their voice. For example, a customer speaking at a high volume for an extended period suggests they are likely experiencing anger or dissatisfaction. However, due to individual differences among customers, some customers may speak loudly without necessarily being angry. In such cases, relying solely on volume and continuity is insufficient to accurately reflect the customer's emotions. Therefore, to address this issue, this application first performs frame processing on the customer's call voice data, resulting in multiple data frame sequences to prevent the voice features from being averaged. Each data frame sequence contains multiple temporally continuous data points, and the division between each data frame sequence is determined by the fluctuation difference between the data points. For example, for data points x1, x2, x3, and x4, if the fluctuation difference between x2 and x3 is greater than a threshold, then x1 and x2 are divided into one data frame sequence, and x3 and x4 into another. Then, the emotional energy of the multiple data frame sequences is calculated using the following formula:

[0055]

[0056] in, For the emotional energy of the nth data frame sequence, For the m-th data point in the n-th data frame sequence, Let be the Hamming weight of the m-th data point in the n-th data frame sequence, and N be the total number of data points in the n-th data frame sequence.

[0057] In the above formula, This can be understood as the amplitude of volume. When emotionally agitated, a person unconsciously raises their voice, increasing the amplitude of the signal in the time domain. This increases emotional energy, consistent with actual human emotional expression. Then, to ensure a smooth transition between data frame sequences, each data point in the sequence is multiplied by its corresponding Hamming weight. The Hamming weight, based on the periodic fluctuations of the cosine function, adjusts the emotional energy of the entire data frame sequence to be periodic, thus preventing abrupt changes between two consecutive data frame sequences. The Hamming weight can be expressed as:

[0058]

[0059] Then, the emotional energy of multiple data frame sequences is normalized to eliminate the influence of differences in the volume of different speakers and generate an emotional intensity fluctuation curve.

[0060] To reduce the impact of background noise, this application, based on the physiological characteristic that the vocal organs exhibit periodic tremors during human speech, analyzes the fluctuation period of the customer's call voice data before performing frame-by-frame processing. Since the tremor frequency of the vocal organs differs at different volumes, multiple continuous and distinct fluctuation period data sequences are obtained. For any given fluctuation period data sequence, multiple sets of period data points are extracted, each set containing multiple period data points, such as... Figure 2 As shown, Figure 2 This is a waveform diagram of customer call voice data provided in an embodiment of this application. Data points a1, a2, a3, a4, and a5 in 'a' belong to a set of periodic data points, with a period interval between two adjacent periodic data points. Then, the standard deviation of multiple periodic data points in the set is calculated as the volatility index of the set. Based on the volatility index, a corresponding optimization strategy is matched from a preset optimization strategy table. Finally, the set of periodic data points is optimized according to the optimization strategy to reduce background noise while preserving emotional features. Specifically, the optimization of the set of periodic data points according to the optimization strategy can be exemplified by: Figure 2As shown in Figure a, for a periodic data point set [a1, a2, a3, a4, a5], if the optimization range of the optimization strategy is one optimization period, then the average of a1 and a3 is calculated to obtain b1, and b1 is used to replace a2. The average of a2 and a4 is calculated to obtain b2, and b2 is used to replace a3. The average of a3 and a5 is calculated to obtain b3, and b3 is used to replace a4, as shown in Figure b. If the optimization range of the optimization strategy is two optimization periods, then the average of a1, a2, a4, and a5 is calculated to obtain c1, and c1 is used to replace a3, as shown in Figure c.

[0061] S102. Perform feature segmentation on the emotion intensity fluctuation map to obtain multiple emotion features.

[0062] In the above steps, based on the current business scenario, the emotion intensity waveform is converted into multiple emotion feature maps, including energy feature maps, frequency feature maps, and fluctuation feature maps. Then, each of the multiple emotion feature maps is segmented to obtain multiple sub-emotion features corresponding to each emotion feature map. Each sub-emotion feature corresponds to a time window. The multiple sub-emotion features corresponding to each of the multiple emotion feature maps are then time-series aligned, and the mean of the time windows is fused to obtain an emotion intensity fluctuation map segmentation strategy. Finally, based on the emotion intensity fluctuation map segmentation strategy, the emotion intensity fluctuation map is segmented to obtain multiple emotion features. The emotion intensity fluctuation map segmentation strategy can also be understood as a time window segmentation strategy. For example, after the mean of the time windows is fused, multiple emotion feature maps will have multiple pre-defined continuous time windows. At this point, the emotion intensity fluctuation map only needs to be segmented based on these multiple pre-defined continuous time windows. This solution transforms a single emotion intensity curve into a structured feature that integrates multiple feature dimensions, preserving the physiological and acoustic details of emotions while better fitting the current customer service scenario, thus providing more reliable feature input for subsequent confidence assessment and satisfaction analysis.

[0063] S103. Conduct a reliability assessment on multiple emotional features to obtain the confidence level of multiple emotional features.

[0064] In the above steps, for each emotional characteristic, this application considers reliability from multiple dimensions, specifically including: stability, consistency, and physiological rationality. Specifically:

[0065] In a customer service scenario, if a customer jokingly expresses anger, it doesn't necessarily mean they are truly angry, but their emotional intensity is high. Therefore, emotional intensity alone cannot accurately determine the customer's true emotional state. Thus, it's necessary to assess the stability of emotional characteristics to verify the authenticity of the customer's emotional expression. Specifically, this application assesses both duration and intensity stability. For duration, the duration of the current emotional characteristic is scored based on the duration of the actual emotional expression. For example, if the duration of actual anger is 2 seconds, and the duration of the current emotional characteristic is 1.8 seconds, the duration score is 1.8 / 2 = 0.9. For intensity stability, the intensity stability of the current emotional characteristic is scored based on the intensity stability of the actual emotional expression. For example, if the intensity of actual anger is long-lasting and doesn't fluctuate wildly, the ratio of the current emotional characteristic's intensity variance to the historical intensity variance of actual anger is calculated to obtain an intensity stability score. Finally, the duration score and the intensity stability score are averaged to obtain a stability score.

[0066] In a customer service scenario, if a customer wants to file a complaint but speaks softly and slowly, analyzing only the emotional intensity might easily misjudge it as a routine business inquiry. Therefore, it is necessary to score the consistency of emotional characteristics across multiple dimensions. Specifically, this application assesses the reliability of consistency across three aspects: emotional intensity energy, emotional intensity frequency, and emotional intensity fluctuation. Of course, the dimensions analyzed will differ depending on the business scenario. For the three dimensions analyzed in this application, the correlation coefficients between them are calculated. For example, the correlation coefficient between emotional intensity energy and emotional intensity frequency is 'a', the correlation coefficient between emotional intensity frequency and emotional intensity fluctuation is 'b', and the correlation coefficient between emotional intensity fluctuation and emotional intensity energy is 'c'. The consistency score is then the average of the three. It should be noted that if the consistency of an emotional characteristic is low, it indicates that the emotion expressed by that characteristic is more complex, and simple analysis algorithms cannot accurately analyze its true business needs, thus its reliability is relatively low.

[0067] In a customer service scenario, if the customer's background noise is significant, the speech contains a lot of interference, making the emotional characteristics appear unreasonable. For example, the physiological frequency of human speech is between 80-500Hz. If the frequency of emotional characteristics is not in this range, it indicates that the reliability of the emotional characteristics is low. In order to assess this reliability, this application scores emotional characteristics based on various reasonable physiological indicators of the human body. If it meets the reasonable physiological indicators, it is a full score of 1 point; if it does not meet the indicators, it is a score of 0. Finally, the total score is calculated, and then the ratio between the total score and the total score is calculated, i.e., 4 (total score) / 5 (total score) = 0.8, to obtain the physiological reasonableness score.

[0068] Finally, based on the reliability weights corresponding to stability, consistency, and physiological rationality, a weighted average of the stability score, consistency score, and physiological rationality score is calculated to obtain the confidence level of the emotional feature.

[0069] S104. Based on the preset reliability threshold, divide multiple emotion features into multiple high-reliability emotion features and multiple low-reliability emotion features.

[0070] In the above steps, the confidence levels of multiple emotion features are compared with preset confidence thresholds. If the confidence level of an emotion feature is greater than or equal to the preset confidence threshold, the emotion feature is determined to be a high-reliability emotion feature. If the confidence level of an emotion feature is less than the preset confidence threshold, the emotion feature is determined to be a low-reliability emotion feature.

[0071] In one possible implementation, since each customer expresses emotions differently, the preset confidence threshold needs to consider each customer's emotional expression habits. Therefore, this application constructs an emotional expression profile of the customer based on customer service call voice data. This profile includes a base frequency range, volume range, and speech rate range. Then, the difference between the emotional expression profile and a preset standard emotional expression profile is calculated to obtain the current customer's confidence adjustment coefficient. The standard emotional expression profile can be understood as a typical emotional expression template, which includes a standard base frequency range, standard volume range, and standard speech rate range. The following method can be used to calculate the difference between the emotional expression profile and the standard emotional expression profile:

[0072] Standard emotional expression profile: standard fundamental frequency is t1, standard fundamental frequency fluctuation range is s1; standard volume range is t2, standard volume fluctuation range is t2; standard speech rate range is t3, standard speech rate fluctuation range is s3.

[0073] Customer emotional expression profile: The actual base frequency is T1, the actual base frequency fluctuation range is S1; the actual volume range is T2, the actual volume fluctuation range is T2; the actual speech rate range is T3, the actual speech rate fluctuation range is T3.

[0074] The difference in fundamental frequency is: ;

[0075] The difference in volume is: ;

[0076] The difference in speech rate is: ;

[0077] At this point, the difference between the standard emotion expression profile and the customer's emotion expression profile... + + ) / 3.

[0078] Then, based on the degree of difference, the confidence adjustment coefficient for the current customer is obtained. ;

[0079] Finally, the confidence threshold of the current customer is obtained by multiplying the current customer's confidence adjustment coefficient by the confidence threshold corresponding to the standard emotion expression profile.

[0080] S105. From the speech-text mapping table, match texts with multiple high-reliability emotion features and texts with multiple low-reliability emotion features.

[0081] S106. Input the text of multiple high-reliability emotional features into the text analysis module to evaluate customer satisfaction and obtain the satisfaction scores corresponding to the multiple high-reliability emotional features.

[0082] In steps S105 to S106 above, under the current business scenario, for high reliability emotion features, the corresponding text can be directly matched according to the speech-text mapping table. For example, the text corresponding to the high reliability emotion feature can be: "Hello", "xx business", "xx product", etc. It should be noted that the text corresponding to the above high reliability emotion feature can also be understood as exclusive high-frequency words of the business scenario. When different customers say the text corresponding to the high reliability emotion feature, their speaking frequency and intensity will not be significantly different. Therefore, there is no need to use the language analysis module to perform speech-to-text operation. It can be directly input into the text analysis module for customer satisfaction evaluation, thereby realizing parallel processing of voice business, greatly reducing the processing pressure of the serial process, improving resource utilization, and improving analysis efficiency.

[0083] For text with low-reliability emotion features, it can be understood as text with high randomness and cannot accurately express the content of the speech. Therefore, it still needs to be input into the speech analysis module for text verification. For text with low-reliability emotion features that pass verification, it can be directly input into the text analysis module for customer satisfaction evaluation. For text with low-reliability emotion features that fail verification, the speech analysis module will re-identify and obtain accurate corrected text before inputting it into the text analysis module for customer satisfaction evaluation. The text verification step in this process can be performed using a lightweight model. At this time, only a small part of the complex text needs to be deeply recognized, which reduces the frequency of calling complex speech recognition models and thus greatly saves speech analysis time.

[0084] Reference Figure 3 This application also provides a real-time analysis system for intelligent business, the system being a server, the server including an acquisition module 1 and an analysis module 2, wherein:

[0085] Module 1 is used to perform time-domain conversion on customer service call voice data to obtain an emotion intensity waveform.

[0086] Analysis module 2 is used to segment the emotion intensity fluctuation map into multiple emotion features; to evaluate the reliability of the multiple emotion features and obtain their confidence levels; based on a preset confidence threshold, the multiple emotion features are divided into multiple high-reliability emotion features and multiple low-reliability emotion features; from the speech-text mapping table, the texts of the multiple high-reliability emotion features and the texts of the multiple low-reliability emotion features are matched; the texts of the multiple high-reliability emotion features are input into the text analysis module to evaluate customer satisfaction and obtain the satisfaction scores corresponding to the multiple high-reliability emotion features.

[0087] In one possible implementation, the customer service call voice data is time-domain converted to obtain an emotion intensity waveform, specifically:

[0088] The customer service call voice data is processed into frames to obtain multiple data frame sequences;

[0089] The emotional energy of multiple data frame sequences is calculated and normalized to obtain the emotional intensity fluctuation curve.

[0090] In one possible implementation, before performing time-domain conversion on the customer service call voice data to obtain the emotion intensity waveform, the method further includes:

[0091] Identify multiple fluctuation periodic data sequences of customer service call voice data, where all data points in a fluctuation periodic data sequence belong to the same periodic function;

[0092] Calculate the volatility index of multiple volatility cycle data sequences;

[0093] Based on the volatility index of multiple volatility cycle data sequences, the corresponding optimization strategies for each of the multiple volatility cycle data sequences are matched from the preset optimization strategy table;

[0094] Based on the optimization strategies corresponding to each of the multiple fluctuation cycle data sequences, the multiple fluctuation cycle data sequences are optimized to obtain optimized customer service call voice data.

[0095] In one possible implementation, the emotion intensity fluctuation graph is segmented to obtain multiple emotion features, specifically including:

[0096] The emotional intensity fluctuation map is converted into multiple emotional feature maps, including an energy feature map, a frequency feature map, and a fluctuation feature map.

[0097] Multiple emotion feature maps are segmented to obtain multiple sub-emotion features corresponding to each emotion feature map, where each sub-emotion feature corresponds to a time window;

[0098] The multiple sub-emotional features corresponding to each of the multiple emotion feature maps are aligned in time sequence, and the mean of the time window is fused to obtain the emotion intensity fluctuation map segmentation strategy.

[0099] Based on the segmentation strategy of the emotion intensity fluctuation map, the emotion intensity fluctuation map is segmented into features to obtain multiple emotion features.

[0100] In one possible implementation, a reliability assessment is performed on multiple emotional features to obtain the confidence levels of the multiple emotional features, specifically as follows:

[0101] Based on multiple preset reliability assessment criteria, multiple emotional characteristics are assessed for reliability, resulting in multiple reliability assessment criteria scores corresponding to each emotional characteristic. These criteria include stability, consistency, and physiological rationality.

[0102] Based on the reliability weights of multiple reliability assessment criteria, the scores of multiple reliability assessment criteria corresponding to each of the multiple emotional features are weighted and averaged to obtain the confidence level of multiple emotional features.

[0103] In one possible implementation, based on a preset reliability threshold, multiple emotion features are divided into multiple high-reliability emotion features and multiple low-reliability emotion features, specifically further including:

[0104] Based on customer service call voice data, a customer emotion expression profile is constructed, which includes the base frequency range, volume range, and speech rate range.

[0105] The confidence adjustment coefficient is obtained by calculating the difference between the customer's emotional expression profile and the preset standard emotional expression profile.

[0106] Based on the confidence adjustment coefficient, the confidence threshold corresponding to the standard emotional expression profile is adjusted to obtain the confidence threshold corresponding to the customer service call voice data.

[0107] In one possible implementation, text containing multiple high-reliability emotional features is input into a text analysis module to perform customer satisfaction evaluation, obtaining satisfaction scores corresponding to the multiple high-reliability emotional features. Specifically, this also includes:

[0108] Multiple low-reliability emotional texts are input into the lightweight speech verification model of the speech analysis module for text verification, resulting in multiple passing texts and multiple failing texts;

[0109] Multiple text inputs are fed into the text analysis module to evaluate customer satisfaction and obtain satisfaction scores corresponding to the multiple texts.

[0110] The speech segments corresponding to multiple failed texts are input into a high-precision speech recognition model to obtain the corrected texts corresponding to each of the multiple failed texts;

[0111] The corrected texts corresponding to the multiple failed texts are input into the text analysis module to evaluate customer satisfaction and obtain the satisfaction scores for the multiple corrected texts.

[0112] It should be noted that the above embodiments of the apparatus are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0113] This application also discloses an electronic device. (See reference...) Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application. The electronic device 400 may include: at least one processor 401, at least one network interface 404, a user interface 403, a memory 405, and at least one communication bus 402.

[0114] The communication bus 402 is used to enable communication between these components.

[0115] The user interface 403 may include a display screen and a camera. Optionally, the user interface 403 may also include a standard wired interface and a wireless interface.

[0116] The network interface 404 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0117] The processor 401 may include one or more processing cores. The processor 401 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 405, and by calling data stored in memory 405. Optionally, the processor 401 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 401 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor 401.

[0118] The memory 405 may include random access memory (RAM) or read-only memory. Optionally, the memory 405 may include a non-transitory computer-readable storage medium. The memory 405 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 405 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 405 may also be at least one storage device located remotely from the aforementioned processor 401. (Refer to...) Figure 4The memory 405, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for real-time analysis of intelligent services.

[0119] exist Figure 4 In the illustrated electronic device 400, the user interface 403 is mainly used to provide an input interface for the user and acquire user input data; while the processor 401 can be used to call an application program stored in the memory 405 for real-time analysis of intelligent services. When executed by one or more processors 401, the electronic device 400 performs one or more of the methods described in the above embodiments. It should be noted that, for the foregoing method embodiments, for the sake of simplicity, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0120] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0121] In the various embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between apparatuses or units may be electrical or other forms.

[0122] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0123] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0124] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, 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. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0125] The above description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Other embodiments of this disclosure will be readily apparent to those skilled in the art upon consideration of the specification and the disclosure of practical truths.

[0126] This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for real-time analysis of intelligent services, characterized in that, Applied to a server, the method comprises: Converting the customer service call voice data into time domain to obtain an emotion intensity waveform; Dividing the emotion intensity waveform into features to obtain a plurality of emotion features, which specifically comprises: Converting the emotion intensity waveform into a plurality of emotion feature maps, wherein the plurality of emotion feature maps include an energy feature map, a frequency feature map, and a fluctuation feature map; Dividing the plurality of emotion feature maps into features to obtain a plurality of sub-emotion features corresponding to each of the plurality of emotion feature maps, wherein each sub-emotion feature corresponds to a time window; Aligning the plurality of sub-emotion features corresponding to each of the plurality of emotion feature maps in time sequence, and performing mean fusion of the time window to obtain an emotion intensity waveform division strategy; According to the emotion intensity waveform division strategy, dividing the emotion intensity waveform into features to obtain a plurality of emotion features; Performing reliability evaluation on the plurality of emotion features to obtain a confidence of the plurality of emotion features; According to a pre-set confidence threshold, dividing the plurality of emotion features into a plurality of high-reliability emotion features and a plurality of low-reliability emotion features; Matching the text of the plurality of high-reliability emotion features and the text of the plurality of low-reliability emotion features from a voice-text mapping table; Inputting the text of the plurality of high-reliability emotion features into a text analysis module to evaluate customer satisfaction to obtain a satisfaction score corresponding to the plurality of high-reliability emotion features.

2. The method of claim 1, wherein, The converting the customer service call voice data into time domain to obtain an emotion intensity waveform specifically comprises: Performing frame processing on the customer service call voice data to obtain a plurality of data frame sequences; Calculating the emotion energy of the plurality of data frame sequences and performing normalization processing to obtain the emotion intensity fluctuation curve.

3. The method of claim 1, wherein, Before the converting the customer service call voice data into time domain to obtain an emotion intensity waveform, further comprising: Identifying a plurality of fluctuation period data sequences of the customer service call voice data, wherein all data points in a fluctuation period data sequence belong to the same periodic function; Calculating the fluctuation index of the plurality of fluctuation period data sequences; According to the fluctuation index of the plurality of fluctuation period data sequences, matching the optimization strategy corresponding to each of the plurality of fluctuation period data sequences from a pre-set optimization strategy table; According to the optimization strategy corresponding to each of the plurality of fluctuation period data sequences, optimizing the plurality of fluctuation period data sequences to obtain optimized customer service call voice data.

4. The method of claim 1, wherein, The performing reliability evaluation on the plurality of emotion features to obtain a confidence of the plurality of emotion features specifically comprises: According to a plurality of pre-set reliability evaluation standards, performing reliability evaluation on the plurality of emotion features to obtain a plurality of reliability evaluation standard scores corresponding to each of the plurality of emotion features, wherein the plurality of reliability evaluation standards include stability, consistency, and physiological rationality; According to the reliability weight of the plurality of reliability evaluation standards, performing weighted average on the plurality of reliability evaluation standard scores corresponding to each of the plurality of emotion features to obtain a confidence of the plurality of emotion features.

5. The method of claim 1, wherein, The multiple emotion features are divided into multiple high-reliability emotion features and multiple low-reliability emotion features according to a preset confidence threshold, and specifically further comprising: According to the customer service call voice data, a customer emotion expression portrait is constructed, and the emotion expression portrait includes a fundamental frequency range, a volume range, and a speech rate range; The customer emotion expression portrait is compared with a preset standard emotion expression portrait to perform difference calculation, and a confidence adjustment coefficient is obtained; According to the confidence adjustment coefficient, a confidence threshold corresponding to the standard emotion expression portrait is adjusted to obtain a confidence threshold corresponding to the customer service call voice data.

6. The method of claim 1, wherein, The text of the multiple high-reliability emotion features is input into a text analysis module to perform customer satisfaction evaluation, and a satisfaction score corresponding to the multiple high-reliability emotion features is obtained, and specifically further comprising: The text of the multiple low-reliability emotion features is input into a lightweight voice verification model of a voice analysis module to perform text verification, and multiple passed texts and multiple failed texts are obtained; The multiple passed texts are input into the text analysis module to perform customer satisfaction evaluation, and a satisfaction score corresponding to the multiple passed texts is obtained; The voice segments corresponding to the multiple failed texts are input into a high-precision voice recognition model to obtain corrected texts corresponding to the multiple failed texts respectively; The corrected texts corresponding to the multiple failed texts are input into the text analysis module to perform customer satisfaction evaluation, and a satisfaction score corresponding to the multiple corrected texts is obtained.

7. A system for real-time analysis of intelligent services for performing a method for real-time analysis of intelligent services according to any one of claims 1 to 6, characterized in that, The system is a server, and the server comprises an acquisition module (1) and an analysis module (2), wherein: The acquisition module (1) is configured to perform time domain conversion on the customer service call voice data to obtain an emotion intensity waveform graph; The analysis module (2) is configured to perform feature division on the emotion intensity waveform graph to obtain multiple emotion features, perform reliability evaluation on the multiple emotion features to obtain confidence of the multiple emotion features, divide the multiple emotion features into multiple high-reliability emotion features and multiple low-reliability emotion features according to a preset confidence threshold, match texts of the multiple high-reliability emotion features and texts of the multiple low-reliability emotion features from a voice-text mapping table, input the texts of the multiple high-reliability emotion features into a text analysis module to perform customer satisfaction evaluation, and obtain a satisfaction score corresponding to the multiple high-reliability emotion features.

8. An electronic device, comprising: The electronic device (400) comprises a processor (401), a memory (405), a user interface (403), and a network interface (404), the memory (405) is configured to store instructions, the user interface (403) and the network interface (404) are configured to communicate with other devices, and the processor (401) is configured to execute the instructions stored in the memory (405) to enable the electronic device (400) to perform the method in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions, and when the instructions are executed, the method in any one of claims 1 to 6 is performed.

Images