An AI-based voice service full-process intelligent quality inspection method and system associated with business work orders

Abstract

The application discloses a voice service full-process intelligent quality inspection method and system based on AI and business work order association, which comprises the following steps: collecting voice signals, associating business work order data, system concurrent data and business term update data, calculating core parameters, dynamically adjusting quality inspection standards based on parameter quantization coupling deviation and standard iteration requirements, and executing quality inspection combined with AI original semantic recognition accuracy and outputting results. The system corresponds to realize the above method. Through core parameter linkage and dynamic standard adjustment, the application solves the quality inspection problems caused by high concurrent coupling deviation and term change, and guarantees the reliability of full-process quality inspection.

Description

Technical Field

[0001] This invention relates to the field of voice service quality inspection technology, specifically to a method and system for intelligent quality inspection of the entire voice service process based on AI and business work order association. Background Technology

[0002] Voice service quality inspection is a crucial step in ensuring service quality. Current technologies largely rely on fixed standards to match AI semantic recognition with business work orders. However, in high-concurrency scenarios, semantic information loss due to bandwidth compression during voice signal transmission, coupled with background noise further interfering with semantic recognition accuracy, and potential delays in the generation of associated business work orders, can lead to inaccurate alignment between voice and work order timestamps, creating implicit coupling bias. Existing solutions cannot quantify this coupling bias, relying solely on static standards for quality inspection, resulting in discrepancies between inspection results and actual service conditions, failing to meet the precise quality inspection requirements of high-concurrency scenarios.

[0003] Based on the above problems, there is an urgent need for a technical solution that can accurately quantify coupling deviations and dynamically adapt to quality inspection standards. Summary of the Invention

[0004] This invention provides an intelligent quality inspection method for the entire process of voice services based on AI and business work orders, comprising the following steps:

[0005] S1: Collect voice signals, associated business work order data, system concurrency data, and business terminology update data during the voice service process. Voice signals include voice bandwidth compression rate, background noise intensity, and voice duration. Associated business work order data includes the completeness of core work order fields, the overlap between work order fields and voice semantics, and work order generation delay time. System concurrency data includes the current system concurrency and the maximum system concurrency. Business terminology update data includes the number of newly added terms, the old terminology obsolescence rate, and the total number of basic terms.

[0006] S2: Calculate the speech semantic distortion coefficient, work order field correlation degree, timestamp misalignment coefficient, terminology change strength, AI terminology recognition confidence, new terminology training sample size, and semantic similarity between new and old terms based on the collected data.

[0007] S3: Calculate the coupling deviation value based on the speech and semantic distortion coefficient, the correlation degree of the work order field and the timestamp misalignment coefficient. Calculate the urgency of standard iteration based on the terminology change strength, AI terminology recognition confidence, the number of newly added terms and the number of training samples for new terms. Combine the coupling deviation value, the current system concurrency, the maximum system concurrency, the urgency of standard iteration and semantic similarity to dynamically adjust the quality inspection standard.

[0008] S4: Perform full-process quality inspection based on the adjusted quality inspection standards and the accuracy of AI's original semantic recognition, and output the quality inspection results.

[0009] Preferably, the speech semantic distortion coefficient is obtained by weighting the speech bandwidth compression rate and the background noise intensity, and the weighting coefficient is preset based on the transmission characteristics of the speech signal.

[0010] Preferably, the correlation degree of work order fields is obtained by weighting the integrity of the core fields of the work order and the overlap between the work order fields and the speech semantics, and the weighting coefficient is based on the preset importance level of the fields of the business work order.

[0011] Preferably, the timestamp misalignment coefficient is calculated by the ratio of the work order generation delay time to the voice duration. During the calculation process, the voice duration is processed to be non-zero to avoid meaningless calculations.

[0012] Preferably, the coupling deviation value is calculated using the voice-work order coupling deviation calculation formula, which is a collaborative quantization expression of the voice semantic distortion coefficient, the work order field correlation degree, and the timestamp misalignment coefficient.

[0013] Preferably, when dynamically adjusting quality inspection standards, it is necessary to calculate the iteration weight of the quality inspection standards. The iteration weight of the quality inspection standards is the result of a joint calculation of the urgency of standard iteration and the semantic similarity between new and old terms.

[0014] Preferably, when dynamically adjusting the quality inspection standards, it is also necessary to calculate the dynamic correction coefficient of the quality inspection standards. When performing full-process quality inspection, the quality inspection result is obtained by combining the final quality inspection accuracy calculation formula. The final quality inspection accuracy calculation formula is a collaborative operation expression of the AI ​​original semantic recognition accuracy, the dynamic correction coefficient of the quality inspection standards, and the coupling deviation value.

[0015] Preferably, the dynamic correction coefficient of the quality inspection standard is calculated by the current system concurrency, the maximum system concurrency, and the voice-work order coupling deviation value. The calculation logic is positively correlated with the growth trend of system concurrency and the magnitude of the coupling deviation value.

[0016] Preferably, an intelligent quality inspection system for the entire process of voice services based on AI and business work orders includes a data acquisition unit, a parameter calculation unit, a standard adjustment unit, and a quality inspection execution unit. The data acquisition unit is used to collect data such as voice bandwidth compression rate, background noise intensity, voice duration, completeness of core fields of the work order, semantic overlap between work order fields and voice, work order generation delay time, current system concurrency, maximum system concurrency, number of new terms, old terminology elimination rate, total number of basic terms, AI terminology recognition confidence, training sample size of new terms, and semantic similarity between new and old terms. The parameter calculation unit is used to calculate the voice semantic distortion coefficient, work order field correlation, timestamp misalignment coefficient, terminology change strength, voice-work order coupling deviation value, standard iteration urgency, quality inspection standard dynamic correction coefficient, and quality inspection standard iteration weight. The standard adjustment unit is used to dynamically adjust the quality inspection standard based on the output of the parameter calculation unit. The quality inspection execution unit is used to perform the entire process quality inspection based on the adjusted quality inspection standard and the accuracy of the original AI semantic recognition and output the quality inspection results.

[0017] Preferably, the data acquisition unit establishes a one-way data transmission connection with the parameter calculation unit, the parameter calculation unit establishes a one-way data transmission connection with the standard adjustment unit, the standard adjustment unit establishes a one-way data transmission connection with the quality inspection execution unit, and all data acquired by the data acquisition unit is synchronously stored in the system's built-in cache module, and the cache module establishes a two-way data interaction connection with the parameter calculation unit.

[0018] The present invention has the following beneficial effects:

[0019] This invention collects multi-dimensional core data to quantify the quality inspection standard adaptation requirements caused by voice-work order coupling deviation and terminology changes, dynamically adjusts the quality inspection standards, and creatively achieves coordinated adaptation of coupling deviation correction and standard iteration. It accurately solves the problem of inaccurate quality inspection caused by coupling deviation in high-concurrency scenarios, and ensures the reliability of quality inspection throughout the entire voice service process. Attached Figure Description

[0020] Figure 1 This is a flowchart of the intelligent quality inspection method for the entire process of voice service based on AI and business work orders in this application;

[0021] Figure 2 This is a connection diagram of the AI-based intelligent quality inspection system for the entire process of voice services linked to business work orders, as described in this application. Detailed Implementation

[0022] 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 embodiments of the present invention, and not all embodiments. 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.

[0023] Technical problems with existing technologies: In high-concurrency scenarios, voice signals are prone to semantic distortion due to bandwidth compression and background noise; the generation delay of related business work orders leads to timestamp misalignment; business terminology changes dynamically with business development; existing solutions cannot quantify the deviations caused by the coupling of multiple factors and rely only on static quality inspection standards, resulting in inaccurate quality inspection results and failing to meet the needs of accurate quality inspection throughout the entire process.

[0024] Based on this, please refer to Figure 1 and Figure 2 This embodiment provides an intelligent quality inspection method for the entire process of voice services based on AI and business work orders, including the following steps:

[0025] S1: Collect voice signals, associated business work order data, system concurrency data, and business terminology update data during the voice service process. Voice signals include voice bandwidth compression rate, background noise intensity, and voice duration. Associated business work order data includes the completeness of core work order fields, the overlap between work order fields and voice semantics, and work order generation delay time. System concurrency data includes the current system concurrency and the maximum system concurrency. Business terminology update data includes the number of newly added terms, the old terminology obsolescence rate, and the total number of basic terms.

[0026] S2: Calculate the speech semantic distortion coefficient, work order field correlation degree, timestamp misalignment coefficient, terminology change strength, AI terminology recognition confidence, new terminology training sample size, and semantic similarity between new and old terms based on the collected data.

[0027] S3: Calculate the coupling deviation value based on the speech and semantic distortion coefficient, the correlation degree of the work order field and the timestamp misalignment coefficient. Calculate the urgency of standard iteration based on the terminology change strength, AI terminology recognition confidence, the number of newly added terms and the number of training samples for new terms. Combine the coupling deviation value, the current system concurrency, the maximum system concurrency, the urgency of standard iteration and semantic similarity to dynamically adjust the quality inspection standard.

[0028] S4: Perform full-process quality inspection based on the adjusted quality inspection standards and the accuracy of AI's original semantic recognition, and output the quality inspection results.

[0029] It is worth mentioning that data acquisition in S1 is achieved through a dedicated data acquisition interface. The voice signal is captured in real time by the voice acquisition device, the voice bandwidth compression rate is obtained by parsing the compression parameters of the voice transmission protocol, the background noise intensity is calculated by the signal-to-noise ratio analysis of the voice signal, and the voice duration is obtained by statistically analyzing the timestamp difference of the voice frames. Related business work order data is extracted through the database interface connected to the business work order system. The integrity of the work order core fields is determined by verifying the non-empty state of the preset core fields. The overlap between the work order fields and the voice semantics is obtained by comparing them with the text semantic matching algorithm. The work order generation delay time is calculated by the difference between the work order creation time and the voice service end time. System concurrency data is obtained through the process monitoring module of the system server. The current system concurrency count is the number of voice service requests processed at the same time. The maximum system concurrency is the preset system processing capacity threshold. Business terminology update data is obtained by monitoring the change log of the business terminology library. The number of newly added terms is the number of newly entered terminology entries within the period. The old terminology elimination rate is calculated by the ratio of the number of old terms deleted within the period to the total number of original terms. The total number of basic terms is the total number of terminology entries in the terminology library at the beginning of the period. The parameter calculations in S2 are performed by a dedicated computing module. The speech semantic distortion coefficient is obtained by weighted summation of speech bandwidth compression rate and background noise intensity. The work order field correlation degree is calculated by weighted fusion of core field integrity and semantic overlap. The timestamp misalignment coefficient is obtained by non-zeroing the ratio of work order generation delay time to speech duration. The terminology change intensity is comprehensively quantified by combining the number of newly added terms, the old terminology elimination rate, and the total number of basic terms. The AI ​​terminology recognition confidence is output by the AI ​​semantic recognition model. The training sample size for new terms is the number of data entries used for training new terms in the AI ​​model. The semantic similarity between new and old terms is calculated by word vector space distance. The coupling deviation value in S3 is calculated by a multi-factor collaborative quantification formula. The standard iteration urgency is derived by combining factors such as terminology change intensity and AI recognition confidence. When dynamically adjusting the quality inspection standards, the standard leniency is adjusted according to the coupling deviation value, and the terminology adaptation range is adjusted according to the standard iteration urgency and semantic similarity. The full-process quality inspection in S4 is carried out through an AI quality inspection model combined with dynamically adjusted standards. The quality inspection process covers the entire stage of voice service access, interaction, and termination. The output quality inspection results include the judgment of whether the quality inspection is qualified and the associated key data identifiers.

[0030] The technical effects achieved by this solution include: comprehensively covering multiple factors that affect the accuracy of quality inspection, realizing dynamic adaptation of quality inspection standards, and ensuring the accuracy and reliability of quality inspection throughout the entire process.

[0031] Technical problems with existing technologies: Existing solutions do not clearly define the calculation logic of speech semantic distortion coefficients, but simply superimpose relevant influencing factors, which makes it impossible to accurately quantify the degree of distortion of speech semantics caused by transmission loss and noise interference, affecting the accuracy of subsequent deviation calculations.

[0032] Based on this, the speech semantic distortion coefficient is calculated by weighting the speech bandwidth compression rate and background noise intensity, with the weighting coefficients preset based on the transmission characteristics of the speech signal. It is worth noting that the speech bandwidth compression rate is the proportion of the speech signal compressed to adapt to the transmission bandwidth; it is dimensionless and ranges from 0 to 1. A higher compression rate indicates a more severe loss of the original semantic information of the speech signal. For example, the compression rate of an uncompressed speech signal is 0, while after heavy compression, the compression rate can reach 0.8. Background noise intensity is the degree to which noise components in the speech signal interfere with the effective semantic information; it is dimensionless and ranges from 0 to 1. It is derived by converting the effective signal power of the speech signal to the noise power. The higher the noise power, the higher the background noise intensity value; for example, the background noise intensity is 0.1 in a quiet environment and can reach 0.9 in a noisy environment. The weighting coefficients are preset based on the transmission medium and protocol of the speech signal. When the transmission medium is optical fiber, the weighting coefficient for speech bandwidth compression rate is set to 0.6, and the weighting coefficient for background noise intensity is set to 0.4, because compression loss has a more significant impact on semantics in optical fiber transmission. When the transmission medium is a wireless network, the weighting coefficient for speech bandwidth compression rate is set to 0.4, and the weighting coefficient for background noise intensity is set to 0.6, because noise interference has a more prominent impact in wireless networks. During calculation, the product of the speech bandwidth compression rate and its corresponding weighting coefficient is added to the product of the background noise intensity and its corresponding weighting coefficient to obtain the speech semantic distortion coefficient. This coefficient is dimensionless and ranges from 0 to 1. A larger value indicates more severe speech semantic distortion, providing a precise quantitative basis for subsequent coupling deviation calculations.

[0033] The technical effects achieved by this solution include: the calculation logic of the speech semantic distortion coefficient is consistent with the actual transmission scenario, the quantification results are accurate, and it effectively supports subsequent deviation quantification and standard adjustment.

[0034] Technical problems with existing technologies: Existing solutions do not clearly define the calculation method for the correlation degree of work order fields, and do not consider the differences in the importance of different fields to semantic matching, which makes it impossible to accurately represent the semantic fit between work orders and voice, affecting the reliability of coupling deviation quantification.

[0035] Based on this, the correlation degree of work order fields is calculated by weighting the completeness of the core fields of the work order and the overlap between the work order fields and the speech semantics. The weighting coefficient is based on the preset importance level of the fields in the business work order.

[0036] It's worth noting that the completeness of core fields in a work order refers to the degree of completeness of fields that play a crucial role in business judgment. It is dimensionless and ranges from 0 to 1. Preset core fields include business type, user requirements, and processing results. It is calculated by the ratio of the number of non-empty and valid core fields to the total number of core fields. For example, if there are 5 core fields and 4 are valid, the completeness is 0.8. The overlap between work order fields and speech / semantic information refers to the degree of overlap between the content described by the work order fields and the speech / semantic information. It is dimensionless and ranges from 0 to 1. It is calculated by segmenting the work order field text and the speech-to-text text, and then calculating the ratio of the number of identical words to the total number of words. For example, if the total number of words in the work order field text and the speech-to-text text is 100, and the number of identical words is 70, the overlap is 0.7. The weighting coefficients are determined based on the importance level of the fields in the business scenario. In the financial business scenario, the weighting coefficient for the completeness of the core fields of the work order is set to 0.7, and the weighting coefficient for the overlap between the work order fields and the semantics of the speech is set to 0.3, because the completeness of the core fields is more critical for quality inspection in financial business. In the e-commerce business scenario, the weighting coefficients for both are set to 0.5, because field completeness and semantic overlap are equally important. During calculation, the product of the completeness of the core fields of the work order with the corresponding weighting coefficient is added to the product of the overlap between the work order fields and the speech semantics with the corresponding weighting coefficient to obtain the work order field relevance. This coefficient is dimensionless and ranges from 0 to 1; a larger value indicates a more solid semantic matching foundation between the work order and the speech.

[0037] The technical effects achieved by this solution include: the calculation of the correlation degree of work order fields takes into account both field integrity and semantic overlap; the weighted logic is in line with actual business; the quantification results are reliable; and it provides strong support for the calculation of coupling deviation.

[0038] Technical problems with existing technologies: The existing solutions do not clearly define the calculation logic of the timestamp misalignment coefficient and do not consider the extreme case of zero voice duration, which may result in invalid calculation results and make it impossible to accurately quantify the degree of interference of timestamp misalignment on quality inspection matching.

[0039] Based on this, the timestamp misalignment coefficient is calculated as the ratio of the work order generation delay time to the voice duration. During the calculation, the voice duration is processed to be non-zero to avoid meaningless calculations. It's worth noting that the work order generation delay time is the time interval between the end of the voice service and the completion of the associated work order generation, measured in seconds. A longer delay time indicates a more severe timestamp misalignment between the voice and the work order. For example, a delay of 1 second after the voice service ends results in a delay of 1, while a delay of 10 seconds results in a delay of 10. The voice duration is the total duration of the voice service from start to finish, measured in seconds. A shorter duration indicates a more significant impact of timestamp misalignment on matching. For example, a 10-second delay has a relatively small impact when the voice duration is 60 seconds, but a 10-second delay has a significant impact when the voice duration is 5 seconds. The non-zero processing is achieved by adding a fixed constant of 1 to the voice duration to avoid the problem of a zero denominator caused by a zero voice duration, ensuring that the calculation result is always valid. During the calculation, the work order generation delay time is divided by (voice duration + 1) to obtain the timestamp misalignment coefficient. This coefficient is dimensionless and ranges from 0 to 1. The larger the value, the more serious the interference of timestamp misalignment on quality inspection matching. For example, when the work order generation delay time is 5 seconds and the voice duration is 9 seconds, the timestamp misalignment coefficient is 5 / (9+1) = 0.5.

[0040] The technical effects achieved by this solution include: rigorous calculation logic for the timestamp misalignment coefficient, effective avoidance of invalid calculations caused by extreme cases, and accurate quantitative results reflecting the degree of interference from timestamp misalignment.

[0041] Technical problems with existing technologies: Existing solutions do not construct a multi-factor collaborative coupling deviation quantification formula, but only quantify a single factor independently, which cannot fully reflect the combined effects of speech semantic distortion, insufficient association of work order fields and timestamp misalignment, resulting in inaccurate coupling deviation quantification.

[0042] Based on this, the coupling deviation value is calculated using the voice-work order coupling deviation calculation formula, which is as follows:

[0043] ;

[0044] It is worth mentioning that the formula is theoretically designed based on the theory of multi-factor coupling effects. The matching deviation between voice and work order is the result of semantic mismatch and temporal misalignment. Quantifying a single factor cannot reflect the cumulative effect of deviations in real-world scenarios. The parameters are defined as follows: This is the voice-work order coupling deviation value, which is dimensionless and ranges from 0 to 100. The larger the value, the more severe the overall deviation. The semantic distortion coefficient is a dimensionless coefficient that ranges from 0 to 1, representing the degree of semantic distortion of the speech signal caused by transmission loss and noise interference. The correlation degree of the work order field is dimensionless, with a value range of 0 to 1, representing the semantic matching basis between the work order and the voice. The timestamp misalignment coefficient is dimensionless, ranging from 0 to 1, and represents the degree of interference caused by timestamp misalignment to the matching. The logical derivation process is as follows: First, With (1- Multiplication is used. Reflecting the destructive effect of semantic distortion, (1- This reflects the weakness of the matching foundation. The product of the two can accurately reflect the synergistic destructive effect of semantic distortion and weak matching, which conforms to the core logic of semantic matching quality = basic matching degree × no distortion coefficient; secondly, Using square operations, because in practice the interference of misalignment on matching does not increase linearly, when Beyond 0.5, the matching error amplifies exponentially. Square operations accurately characterize this non-linearity, avoiding excessive weighting for low misalignment and insufficient weighting for high misalignment. Then, square root operations normalize the superposition of the semantic collaboration disruption effect and the temporal misalignment non-linearity effect, ensuring the result within the square root falls between 0 and 2, and the result after the square root falls between 0 and √2. Multiplying by 100 maps the value to the range of 0 to 141.4. After verification in real-world scenarios, 0 to 100 is taken as the valid interval, ensuring both quantification accuracy and easy intuitive judgment of the deviation level. Finally, all parameters are designed as dimensionless coefficients, ensuring complete dimensional consistency and no unit conflicts during calculation, conforming to the basic principles of mathematical operations. The calculation process is executed in real-time by the system's computing unit. Based on the collected raw data and calculated intermediate parameters, the coupling deviation value can be obtained by substituting them into the formula, providing a precise basis for the dynamic adjustment of quality inspection standards.

[0045] The technical effects achieved by this solution include: the coupling deviation value can comprehensively and accurately reflect the combined influence of multiple factors, the quantitative results are intuitive and reliable, and it provides scientific data support for subsequent standard adjustments.

[0046] Technical problems with existing technologies: Existing solutions do not clearly define the linkage calculation logic of the iteration weight of quality inspection standards, and cannot balance the adaptation needs of terminology changes with the stability of quality inspection standards, resulting in either excessive or insufficient iteration, which affects the accuracy of quality inspection in terminology change scenarios.

[0047] Therefore, when dynamically adjusting quality inspection standards, it is necessary to calculate the iteration weight of the quality inspection standards. The formula for calculating the iteration weight of the quality inspection standards is as follows:

[0048] ;

[0049] It's worth noting that the formula's theoretical design is based on the iterative requirements-semantic coherence balance theory. The core of standard iteration is finding a balance between rapidly adapting to new terminology and maintaining standard stability. Excessive iteration leads to frequent standard changes, affecting quality inspection consistency, while insufficient iteration fails to adapt to changes in business terminology. The parameters are defined as follows: This is the iteration weight of the quality inspection standard, dimensionless, with a value ranging from 0.3 to 0.8. The larger the value, the greater the standard iteration range and the higher the degree of integration of new terms. To assess the urgency of standard iteration, the value is dimensionless and ranges from 0 to 120. It is calculated by combining the intensity of term change, the confidence of AI term recognition, the number of newly added terms, and the training sample size of new terms. The larger the value, the more urgent the need for standard iteration. The semantic similarity between new and old terms is dimensionless, ranging from 0 to 1. It is obtained by comparing the textual features of the new and old terms using a semantic matching algorithm. Higher similarity indicates stronger compatibility between the new term and the existing standard. The logical derivation process is as follows: First, a base weight of 0.5 is set to ensure that even in scenarios with the lowest iteration urgency and highest semantic similarity, the iteration weight remains close to 0.5, guaranteeing that new terms have basic room for integration and avoiding standard rigidity; Second, For iterative demand adjustment items, because The maximum value is 120, and the adjustment term ranges from 0 to 0.5. The higher the iteration urgency, the larger the adjustment term value, and the closer the iteration weight is to 0.8, meeting the rapid adaptation requirements in high-change scenarios. The denominator design of 240 ensures smooth adjustment amplitude and avoids abrupt changes; then, As a semantic coherence constraint, because The value ranges from 0 to 1, while the constraint term ranges from 0 to 0.005. Higher semantic similarity results in a larger constraint term value, and the iteration weight is closer to 0.5, avoiding excessive iteration on highly similar terms and ensuring the consistency of the standard. The denominator of 200 controls the influence of the constraint term, ensuring that iteration requirements remain the core dominant factor. Finally, after the three factors are combined... The value range is strictly controlled between 0.3 and 0.8, which satisfies the iteration requirements of different scenarios while ensuring the stability of the standard. During calculation, the computing unit calculates based on the already calculated... and Substituting these values ​​into the formula yields the iteration weights of the quality inspection standards, which are used to determine the update range of terminology adaptation in the quality inspection standards.

[0050] The technical effects achieved by this solution include: the iterative weighting of quality inspection standards takes into account both iterative needs and semantic coherence, the quantification results are reasonable, and the scientific nature and stability of standard iteration are guaranteed.

[0051] Technical problems with existing technologies: Existing solutions do not clearly define the collaborative calculation formula for the final quality inspection accuracy, and cannot effectively integrate the original AI recognition results, dynamic correction coefficients and coupling deviation values, resulting in quality inspection results that cannot adapt to scenario deviations and have insufficient accuracy.

[0052] Therefore, when dynamically adjusting quality inspection standards, it is also necessary to calculate the dynamic correction coefficient of the quality inspection standards. When performing full-process quality inspection, the quality inspection result is obtained by combining it with the final quality inspection accuracy calculation formula. The final quality inspection accuracy calculation formula is as follows:

[0053] ;

[0054] It's worth noting that the formula's theoretical design is based on the core identification-deviation compensation collaborative theory. The core of quality inspection accuracy lies in AI's semantic recognition capability, but this requires dynamically adjusting coefficients to adapt to scenario deviations, while simultaneously reducing the risk of result distortion in high-deviation scenarios through deviation compensation terms. The parameters are defined as follows: To determine the final quality inspection accuracy, the unit is percentage, and the value ranges from 0 to 100. The larger the value, the more accurate the quality inspection result. The AI ​​raw semantic recognition accuracy is expressed as a percentage and is directly output by the AI ​​quality inspection model. It represents the model's basic recognition ability for speech semantics and has a value range of 0 to 100. The dynamic correction coefficient for the quality inspection standard is dimensionless and ranges from 0.1 to 1. It is calculated from the current system concurrency, the maximum system concurrency, and the voice-work order coupling deviation value, and represents the dynamic leniency of the quality inspection standard. This is the voice-work order coupling deviation value, dimensionless, ranging from 0 to 100, which is the coupling deviation quantification result mentioned above. The logical derivation process is as follows: First, Constituting the core accuracy items, As a basic indicator of recognition capability, multiplied by Then, dynamic adjustments are made to the basic capabilities and scene adaptation, especially in high-concurrency and high-bias scenarios. As the accuracy is reduced, the core accuracy parameters are adjusted accordingly to avoid misjudgments caused by fixed standards, thus meeting the core requirement that quality inspection standards should dynamically adapt to different scenarios; secondly, This constitutes a deviation compensation item. The coupling deviation value is normalized to the range of 0 to 0.5. (1 - normalized deviation value) represents the inverse coefficient of the deviation's impact on the result. The larger the deviation, the smaller the inverse coefficient, and the smaller the compensation term value. This avoids over-reliance on the compensation term in high-deviation scenarios, which could lead to accuracy distortion. A weighting coefficient of 0.3 ensures that the compensation term only plays an auxiliary corrective role; the core still relies on the adaptation results of AI recognition and dynamic standards. Then, regarding the dimensional balance design... Percentage units and The dimensionless product still has the dimension of a percentage. In the compensation term (1- The value is dimensionless; multiplying it by 0.3 maps it to a percentage, thus unifying the dimensions with the core accuracy item. The output is in percentage form, which meets the need for an intuitive expression of quality inspection accuracy; finally, The denominator of 200 is designed to match the variation of the compensation term with the core accuracy term. The compensation term is 30% when there is no deviation and 15% when there is maximum deviation, ensuring optimized results in unbiased scenarios while avoiding overcompensation in high-deviation scenarios. The calculation process is completed by the quality inspection execution unit; substituting the relevant parameters yields the final quality inspection accuracy, which serves as the core indicator of the quality inspection results.

[0055] The technical effects achieved by this solution include: rigorous calculation logic for final quality inspection accuracy, taking into account both AI recognition capabilities and the impact of dynamic deviations, resulting in accurate and reliable results, and effectively improving the accuracy of quality inspection.

[0056] Technical problems with existing technologies: The existing solutions do not clearly define the relationship between the dynamic correction coefficient of the quality inspection standard and the system concurrency and coupling deviation value, which leads to the risk of deviation in the dynamic correction coefficient not being able to accurately adapt to the scenario. In high concurrency and high deviation scenarios, the standard correction is not timely, which affects the accuracy of the quality inspection results.

[0057] Based on this, the dynamic correction coefficient of the quality inspection standard is calculated using the current system concurrency, the maximum system concurrency, and the voice-work order coupling deviation value. The calculation logic is positively correlated with the growth trend of system concurrency and the magnitude of the coupling deviation value. It's worth noting that the current system concurrency is the number of voice service requests processed simultaneously by the system at a given moment, measured in messages per second. This is statistically analyzed in real time by the system server's process monitoring module. For example, if the system processes 50 voice service requests simultaneously at a given moment, the current system concurrency is 50 messages per second. The maximum system concurrency is the maximum processing capacity threshold designed for the system, also measured in messages per second. It is preset based on the system's hardware configuration and software optimization level; for example, the maximum system concurrency is 100 messages per second. The ratio of these two values ​​constitutes the concurrency load coefficient, which is dimensionless and ranges from 0 to 1. A larger ratio indicates a higher system load and a higher risk of deviation in voice transmission and work order processing. For example, when the current system concurrency is 80 messages per second and the maximum system concurrency is 100 messages per second, the concurrency load coefficient is 0.8. The voice-work order coupling deviation value is the comprehensive deviation quantification result mentioned above. It is dimensionless and ranges from 0 to 100, with a larger value indicating a more severe scenario deviation. The calculation logic is designed as follows: Dynamic correction coefficient for quality inspection standards = 1 - (concurrency load coefficient × 0.5 + coupling deviation value ÷ 100 × 0.5). This logic makes the correction coefficient positively correlated with the system concurrency and coupling deviation value. That is, the higher the system concurrency and the larger the coupling deviation value, the smaller the correction coefficient and the more lenient the quality inspection standard; conversely, the larger the correction coefficient and the stricter the quality inspection standard. During the calculation process, this logic integrates the influence of the concurrency load coefficient and the coupling deviation value to ensure that the dynamic correction coefficient ultimately falls within a reasonable range of 0.1 to 1. This avoids quality inspection misjudgments caused by insufficient correction and prevents standard invalidation caused by overcorrection. For example, when the concurrency load factor is 0.8 and the coupling deviation value is 80, the dynamic correction factor = 1 - (0.8 × 0.5 + 80 ÷ 100 × 0.5) = 1 - (0.4 + 0.4) = 0.2, and the quality inspection standard is appropriately lenient; when the concurrency load factor is 0.2 and the coupling deviation value is 20, the dynamic correction factor = 1 - (0.2 × 0.5 + 20 ÷ 100 × 0.5) = 1 - (0.1 + 0.1) = 0.8, and the quality inspection standard is relatively strict.

[0058] The technical effects achieved by this solution include: the dynamic correction coefficient of the quality inspection standard accurately adapts to the risk of scenario deviation, the quantitative results are reasonable, and the adjustment of the quality inspection standard is timely and targeted.

[0059] Technical problems with existing technologies: The existing system does not clearly define the specific division of labor of each functional unit, the scope of data collection and the core calculation content. The module interaction logic is chaotic and the data transmission path is unclear. It cannot effectively support the collaborative implementation of intelligent quality inspection throughout the entire process, resulting in low quality inspection efficiency and inaccurate results.

[0060] Based on this, this embodiment provides an intelligent quality inspection system for the entire process of voice services based on AI and business work orders. It includes a data acquisition unit, a parameter calculation unit, a standard adjustment unit, and a quality inspection execution unit. The data acquisition unit is used to collect data such as voice bandwidth compression rate, background noise intensity, voice duration, completeness of core work order fields, semantic overlap between work order fields and voice, work order generation delay time, current system concurrency, maximum system concurrency, number of newly added terms, old terminology elimination rate, total number of basic terms, AI terminology recognition confidence, training sample size for new terms, and semantic similarity between new and old terms. The parameter calculation unit is used to calculate the voice semantic distortion coefficient, work order field correlation, timestamp misalignment coefficient, terminology transition strength, voice-work order coupling deviation value, standard iteration urgency, dynamic correction coefficient of quality inspection standards, and iteration weight of quality inspection standards. The standard adjustment unit is used to dynamically adjust the quality inspection standards based on the output of the parameter calculation unit. The quality inspection execution unit is used to perform full-process quality inspection based on the adjusted quality inspection standards and the accuracy of the original AI semantic recognition, and output the quality inspection results.

[0061] It is worth mentioning that the data acquisition unit, as the core of the system's data input, connects to the voice acquisition device, the business work order system, the system server, and the business terminology database through multiple dedicated acquisition interfaces. This ensures the comprehensiveness and relevance of the collected data. All collected data is stored in a standardized format for easy access by the subsequent parameter calculation unit. The parameter calculation unit receives the output from the data acquisition unit and has built-in preset calculation logic and algorithm models, including weighted calculation logic for voice semantic distortion coefficients, weighted calculation logic for work order field correlation, ratio calculation logic for timestamp misalignment coefficients, as well as formulas for calculating voice-work order coupling deviation, iterative weight calculation formulas for quality inspection standards, and final quality inspection accuracy calculation formulas. Through these logics and formulas, the raw data is transformed into core parameters with quantitative significance, realizing the transformation from raw data to decision-making basis. The standard adjustment unit, as the system's decision-making core, receives the coupling deviation value, standard iteration urgency, dynamic correction coefficient of the quality inspection standard, and iteration weight of the quality inspection standard from the parameter calculation unit. Based on preset adjustment rules, it dynamically adjusts the stringency and terminology adaptation range of the quality inspection standard. For example, when the coupling deviation value is high, the stringency of the quality inspection standard is reduced; when the standard iteration urgency is high, the terminology adaptation range is increased, ensuring that the standard always adapts to changes in the actual scenario. The quality inspection execution unit, as the system's execution core, has a built-in AI quality inspection model. It combines the dynamic standard output by the standard adjustment unit with the accuracy of the AI's original semantic recognition, substitutes it into the final quality inspection accuracy calculation formula, completes the entire process of quality inspection, and outputs the quality inspection results, forming a closed-loop logic of data acquisition, parameter calculation, standard adjustment, and quality inspection execution. Fixed data transmission paths are established between the units via a high-speed data bus. The data acquisition unit only transmits data to the parameter calculation unit, the parameter calculation unit only transmits calculation results to the standard adjustment unit, and the standard adjustment unit only transmits the adjusted standard to the quality inspection execution unit, ensuring that each unit functions independently, interacts in an orderly manner, and collaboratively supports the entire process of intelligent quality inspection. The technical effects achieved by this solution include: clear division of labor among system units and clear interaction logic, which can fully support dynamic quality inspection needs and ensure smooth implementation of the quality inspection process and accurate results.

[0062] Technical problems with existing technologies: The existing system does not clearly define the data transmission relationship and data storage mechanism of each unit, the data transmission direction is chaotic, and there is a lack of dedicated data storage modules, which leads to data transmission delays, loss or redundant interaction, affecting the efficiency and stability of quality inspection, and the parameter calculation unit cannot easily call up historical data.

[0063] Based on this, the data acquisition unit establishes a one-way data transmission connection with the parameter calculation unit, the parameter calculation unit establishes a one-way data transmission connection with the standard adjustment unit, and the standard adjustment unit establishes a one-way data transmission connection with the quality inspection execution unit. All data collected by the data acquisition unit is synchronously stored in the system's built-in cache module, and the cache module establishes a bidirectional data interaction connection with the parameter calculation unit. It is worth noting that the one-way data transmission connection is implemented using a one-way data flow channel based on the TCP protocol. Each connection has an independent communication port, clearly defining the sender and receiver of data transmission and prohibiting reverse data flow. The communication port of the data acquisition unit acts only as the sender, and the corresponding port of the parameter calculation unit acts only as the receiver. This avoids data crosstalk and reverse interference at the communication protocol level, ensuring the functional independence of each unit. The cache module adopts a ring buffer storage architecture. The raw data collected by the data acquisition unit is written to the buffer in real time according to the timestamp order. The buffer capacity is preset based on the system's maximum concurrency and data acquisition frequency, ensuring that it can store complete data for at least one business cycle. A first-in, first-out (FIFO) eviction mechanism is also used to avoid excessive resource consumption due to storage redundancy. The bidirectional data interaction between the parameter calculation unit and the caching module is achieved through a dedicated data call interface. During the calculation process, the parameter calculation unit can submit historical data query requests through the interface, specifying query conditions such as time range and data type. The caching module quickly returns the corresponding data based on the request, without needing to be forwarded by the data acquisition unit, significantly improving the efficiency of historical data retrieval. For example, when calculating the intensity of term change, the parameter calculation unit needs to obtain data such as the number of newly added terms and the total number of basic terms in the historical period. This data can be read directly from the caching module through the bidirectional interaction interface, without the need for re-collection, reducing data transmission latency and lowering the load on the data acquisition unit. The data transmission and interaction of each unit are uniformly scheduled by the system's communication control module, ensuring the real-time performance and reliability of data transmission. The data synchronization update frequency of the caching module is consistent with the acquisition frequency of the data acquisition unit, ensuring the consistency between the historical data and real-time data retrieved by the parameter calculation unit.

[0064] The technical effects achieved by this solution include: no crosstalk or reverse interference in system data transmission, significantly improved transmission efficiency, efficient storage and rapid response of the caching module to historical data retrieval requests, providing convenient data support for parameter calculation, reducing the interaction load of each unit, and improving the overall stability of system operation and the processing efficiency of the quality inspection process.

[0065] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent quality inspection of the entire voice service process based on AI and business work order association, characterized in that, Includes the following steps: S1: Collect voice signals, associated business work order data, system concurrency data, and business terminology update data during the voice service process. Voice signals include voice bandwidth compression rate, background noise intensity, and voice duration. Associated business work order data includes the completeness of core work order fields, the overlap between work order fields and voice semantics, and work order generation delay time. System concurrency data includes the current system concurrency and the maximum system concurrency. Business terminology update data includes the number of newly added terms, the old terminology obsolescence rate, and the total number of basic terms. S2: Calculate the speech semantic distortion coefficient, work order field correlation degree, timestamp misalignment coefficient, terminology change strength, AI terminology recognition confidence, new terminology training sample size, and semantic similarity between new and old terms based on the collected data. S3: Calculate the coupling deviation value based on the speech and semantic distortion coefficient, the correlation degree of the work order field and the timestamp misalignment coefficient. Calculate the urgency of standard iteration based on the terminology change strength, AI terminology recognition confidence, the number of newly added terms and the number of training samples for new terms. Combine the coupling deviation value, the current system concurrency, the maximum system concurrency, the urgency of standard iteration and semantic similarity to dynamically adjust the quality inspection standard. S4: Perform full-process quality inspection based on the adjusted quality inspection standards and the accuracy of AI's original semantic recognition, and output the quality inspection results.

2. The intelligent quality inspection method for the entire voice service process based on AI and business work order association as described in claim 1, characterized in that, The speech semantic distortion coefficient is calculated by weighting the speech bandwidth compression rate and the background noise intensity. The weighting coefficient is preset based on the transmission characteristics of the speech signal.

3. The intelligent quality inspection method for the entire voice service process based on AI and business work order association as described in claim 1, characterized in that, The correlation degree of work order fields is calculated by weighting the completeness of the core fields of the work order and the overlap between the work order fields and the speech semantics. The weighting coefficient is based on the preset importance level of the fields in the business work order.

4. The intelligent quality inspection method for the entire voice service process based on AI and business work order association as described in claim 1, characterized in that, The timestamp misalignment coefficient is calculated by the ratio of the work order generation delay time to the voice duration. During the calculation process, the voice duration is processed to be non-zero to avoid meaningless calculations.

5. The intelligent quality inspection method for the entire voice service process based on AI and business work order association as described in claim 1, characterized in that, The coupling deviation value is calculated using the voice-work order coupling deviation calculation formula, which is a collaborative quantization expression of the voice semantic distortion coefficient, the work order field correlation degree, and the timestamp misalignment coefficient.

6. The intelligent quality inspection method for the entire voice service process based on AI and business work order association as described in claim 1, characterized in that, When dynamically adjusting quality inspection standards, it is necessary to calculate the iteration weight of the quality inspection standards. The iteration weight of the quality inspection standards is the result of a joint calculation of the urgency of standard iteration and the semantic similarity between new and old terms.

7. The intelligent quality inspection method for the entire voice service process based on AI and business work order association as described in claim 1, characterized in that, When dynamically adjusting quality inspection standards, it is also necessary to calculate the dynamic correction coefficient of the quality inspection standards. When performing full-process quality inspection, the quality inspection result is obtained by combining the final quality inspection accuracy calculation formula. The final quality inspection accuracy calculation formula is a collaborative operation expression of AI original semantic recognition accuracy, quality inspection standard dynamic correction coefficient and coupling deviation value.

8. The intelligent quality inspection method for the entire voice service process based on AI and business work order association as described in claim 7, characterized in that, The dynamic correction coefficient of the quality inspection standard is calculated by the current system concurrency, the maximum system concurrency, and the voice-work order coupling deviation value. The calculation logic is positively correlated with the growth trend of system concurrency and the magnitude of the coupling deviation value.

9. A voice service end-to-end intelligent quality inspection system based on AI and business work order association, characterized in that, The system includes a data acquisition unit, a parameter calculation unit, a standard adjustment unit, and a quality inspection execution unit. The data acquisition unit is used to collect data such as voice bandwidth compression rate, background noise intensity, voice duration, completeness of core fields of work orders, overlap between work order fields and voice semantics, work order generation delay time, current system concurrency, maximum system concurrency, number of new terms, old term obsolescence rate, total number of basic terms, AI term recognition confidence, training sample size of new terms, and semantic similarity between new and old terms. The parameter calculation unit is used to calculate voice semantic distortion coefficient, work order field correlation, timestamp misalignment coefficient, terminology transition strength, voice-work order coupling deviation value, standard iteration urgency, quality inspection standard dynamic correction coefficient, and quality inspection standard iteration weight. The standard adjustment unit is used to dynamically adjust the quality inspection standard based on the output of the parameter calculation unit. The quality inspection execution unit is used to perform full-process quality inspection based on the adjusted quality inspection standard and the accuracy of AI original semantic recognition and output the quality inspection results.

10. The intelligent quality inspection system for the entire voice service process based on AI and business work order association as described in claim 9, characterized in that, The data acquisition unit establishes a one-way data transmission connection with the parameter calculation unit, the parameter calculation unit establishes a one-way data transmission connection with the standard adjustment unit, and the standard adjustment unit establishes a one-way data transmission connection with the quality inspection execution unit. All data collected by the data acquisition unit is synchronously stored in the system's built-in cache module, and the cache module establishes a two-way data interaction connection with the parameter calculation unit.

Images