Multi-platform data interaction system based on voice interaction

Abstract

The invention discloses a multi-platform data interaction system based on voice interaction, and relates to the technical field of voice quality inspection.The multi-platform data interaction system comprises the steps that historical voice files communicating with different seats are obtained through a client id, a communication network stability score S is calculated through an information entropy weighting method, and Sgt is screened out; ; obtaining a primary quality inspection attribute, a secondary quality inspection attribute I, a secondary quality inspection attribute II, a controllable quality inspection attribute and an uncontrollable quality inspection attribute according to the voice file of the voice file to be subjected to quality inspection according to Ggt; ; a and Gamp; lt; if Ggt is equal to a, carrying out classification; ; a, if yes, quality inspection is carried out according to the primary quality inspection attribute and the secondary quality inspection attribute II, and the uncontrollable quality inspection attribute is interspersed between the primary quality inspection attribute and the secondary quality inspection attribute II; lt; if yes, quality inspection is carried out according to the primary quality inspection attribute and the secondary quality inspection attribute I, and the uncontrollable quality inspection attribute is interspersed between the primary quality inspection attribute and the secondary quality inspection attribute I. According to the invention, the key basic attributes in the voice file to be subjected to quality inspection are extracted, and the speed and efficiency of voice quality inspection are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of voice quality inspection, in particular to a multi-platform data interaction system based on voice interaction. BACKGROUND

[0002] With the rapid development of diversified scenarios such as Internet of Things, intelligent cockpit, smart home, etc., voice interaction, as one of the most natural human-computer interaction methods, has been widely used in vehicle terminal, intelligent wear, home control, etc. In a complete service experience, users often need to interact with devices of different platforms through voice, thereby generating cross-platform voice interaction data flow.

[0003] However, the existing voice processing system is usually designed and optimized for a single platform or isolated scenario, lacking unified fusion and quality evaluation mechanism for cross-platform voice interaction data. This leads to a series of problems: first, due to the differences in network environment, device performance and communication protocol, the stability and continuity of cross-platform interaction are difficult to guarantee, affecting user experience; second, it is difficult to comprehensively analyze the interaction behavior and quality of users on multiple platforms from a global perspective, making it difficult to accurately locate service bottlenecks and optimization direction; finally, traditional quality inspection methods mostly focus on post-inspection, which cannot realize real-time perception of network state and interaction quality during the interaction process, and dynamically adjust the interaction logic, resulting in incomplete service closed loop.

[0004] Therefore, there is an urgent need in the art for a system that can break through the barriers of multi-platform data, realize unified quality monitoring and intelligent analysis of cross-scene voice interaction, in order to improve the reliability and user satisfaction of voice interaction in complex environment. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a multi-platform data interaction system based on voice interaction, which solves the problems of network interference, low efficiency of manual work and inaccurate feature analysis in the prior art voice quality inspection technology.

[0006] To achieve the above purpose, the present application realizes the following technical scheme: a multi-platform data interaction system based on voice interaction, comprising: According to the customer id, the historical voice files communicated with different agents are obtained, the communication network stability score S is calculated by using information entropy weighting method, the voice files with S>Sth are screened out, and the primary quality inspection attribute, the secondary quality inspection attribute one, the secondary quality inspection attribute two, the controllable quality inspection attribute and the uncontrollable quality inspection attribute are obtained, wherein the historical voice files contain customer scores and quality inspection conditions; Based on the customer ID, locate the voice file to be inspected. If G>a, check the primary quality inspection attribute. If it meets the requirement, proceed to the secondary quality inspection attribute. If it meets the requirement, the quality inspection is qualified. If the primary quality inspection attribute does not meet the requirement, but it contains an uncontrollable quality inspection attribute and only that attribute is not within the specified threshold, proceed to the secondary quality inspection attribute. If it meets the requirement, the quality inspection is qualified; otherwise, the quality inspection is unqualified. If the primary quality inspection attribute does not meet the requirement except for the uncontrollable quality inspection attribute, the quality inspection is directly judged as unqualified. Here, G is the agent score and a is the scoring threshold. If G <= a, check the primary quality inspection attribute. If it does not meet the requirements, proceed to secondary quality inspection attribute one. If any basic attribute is outside the specified threshold, the quality inspection fails. If there is an uncontrollable quality inspection attribute among the primary quality inspection attributes and only that basic attribute is outside the specified threshold, proceed to secondary quality inspection attribute two. If both are within the specified threshold range, the quality inspection passes. If the primary quality inspection attribute meets the requirements, proceed to secondary quality inspection attribute two. If it meets the requirements, the quality inspection passes; otherwise, the quality inspection fails.

[0007] As a further aspect of the present invention, the specific steps for calculating the stability score S of a communication network using the information entropy weighting method are as follows: Collect network data from historical voice files, specifically signal strength, packet loss rate, latency, and jitter. According to the formula The signal strength is normalized according to the formula. Normalize the packet loss rate, latency, and jitter. For each index i, calculate its probability in all samples. Calculate the information entropy Where n is the number of samples; According to the formula Calculate the weight of each indicator, where m is the number of indicators; According to the formula Calculate the network stability score.

[0008] As a further aspect of the present invention, the basic attributes of voice files with G<=a and failing quality inspection and those with G>a and passing quality inspection are calculated. The basic attributes include volume, speech rate, professionalism, attitude, and timbre.

[0009] As a further aspect of the present invention, the specific steps for determining whether the volume, speaking speed, professionalism, attitude, and timbre are within the specified thresholds are as follows: for volume, count the number of time points num1 where the sound pressure is not within the specified range and the total number of time points num2 on the speech time axis, calculate sound=num1 / num2, and if sound∈[Num1,Num2], then determine that the volume of the seats is consistent, where Num1 and Num2 are the set ratio thresholds; Speech rate: Count the number of pauses fre, the duration of pauses ti, and the total duration T of the agent's speech. Obtain the actual pronunciation duration (T - sum(ti)), and calculate the speech rate spe = word / (T - sum(ti)). If spe ∈ [spe1, spe2], it indicates that the speech rate is appropriate. Here, word is the number of characters or words in the text content, and spe1 and spe2 are the set speech rate thresholds. Professionalism: Record the total number of agent responses Ans and the number of customer follow-up questions ans, calculate the customer follow-up rate pro = ans / Ans, count the number of professional words Np and the total number of words Nt, calculate the professional word frequency per response PWFi = Np / Nt, and take the average value of PWFi for all follow-up responses as the agent's comprehensive SPWF. When pro <= proth, directly determine that the agent is professional. When pro > proth, calculate SS = pro * (1 - SPWF). If SS < Sth, it is determined to be professional; otherwise, it is unprofessional. Here, proth is the set follow-up rate threshold, and Sth is the set weighted score threshold. Attitude: Collect the number of polite words Pos, the number of negative words Neg, the total number of words Total, the number of responses res within a specified time, and the total number of interactions Res. Calculate the text politeness index TPI = (Pos - Neg) / Total, and at the same time calculate the response timeliness RT = res / Res. When TPI >= T1 and RTI >= T2, it is determined that the agent has a good attitude; otherwise, the attitude is poor. Here, T1 and T2 are the set thresholds. Timbre: Obtain the spectral centroid SC. When SC ∈ [SCmin, SCmax], it is determined that the timbre is good; otherwise, the timbre is poor. SCmin and SCmax are the set spectral centroid ranges.

[0010] As a further solution of the present invention, record the number of times aa that each basic attribute of the voice file with G <= a and unqualified quality inspection is not within the threshold range. If aa > [0.8 * AA], classify it as key attribute one. Record the number of times bb that each basic attribute of the voice file with G > a and qualified quality inspection is within the threshold range. If bb > [0.8 * BB], classify it as key attribute two. Here, AA is the total number of recording files with unqualified and poor scores, and BB is the total number of recording files with qualified and excellent scores. [] represents the rounding symbol.

[0011] As a further solution of the present invention, find the intersection of key attribute one and key attribute two and use it as the primary quality inspection attribute. Find the difference set between key attribute one and key attribute two and use it as the secondary quality inspection attribute one. Find the difference set between key attribute two and key attribute one and use it as the secondary quality inspection attribute two.

[0012] As a further aspect of the present invention, the controllable quality inspection attributes include volume, speaking speed, professionalism, and attitude, while the uncontrollable quality inspection attributes include timbre.

[0013] As a further aspect of the present invention, when the primary quality inspection attribute is an empty set, if G>a, then the process switches to secondary quality inspection attribute two. If all attributes meet the threshold range, the quality inspection is qualified; otherwise, the quality inspection is unqualified. If G<=a, then the process switches to secondary quality inspection attribute one. If any attribute is outside the threshold range, the quality inspection is unqualified. If the attribute outside the threshold range is an uncontrollable quality inspection attribute and all other basic attributes are controllable, then the process switches to secondary quality inspection attribute one. If all attributes meet the threshold range, the quality inspection is qualified.

[0014] This invention provides a multi-platform data interaction system based on voice interaction, which has the following advantages compared with the prior art: (1) Unified quality inspection and fusion analysis of cross-platform interactive data has been realized: This invention seamlessly associates the historical voice files generated by users on different platforms such as in-vehicle, home, and mobile terminals through customer ID, and uses the information entropy weighting method to comprehensively evaluate the stability of the communication network, effectively eliminating quality interference caused by platform switching or network fluctuations, and laying a reliable foundation for subsequent accurate quality inspection.

[0015] (2) A closed loop for intelligent quality inspection and optimization for multi-platform interaction has been constructed: This invention extracts key features such as "primary quality inspection attributes" and "secondary quality inspection attributes" from massive interaction data, and distinguishes between "controllable" and "uncontrollable" factors, forming a set of efficient and configurable quality inspection strategies. This not only greatly improves the automation level and efficiency of cross-platform voice quality inspection, but also generates detailed interaction quality analysis reports, providing accurate data support for agent training, interaction process optimization, and even software and hardware improvements on different platforms (such as intelligent cockpit systems and intelligent customer service platforms), ultimately improving overall service quality and user satisfaction. Attached Figure Description

[0016] Figure 1 This is a flowchart of the steps of the present invention. Detailed Implementation

[0017] 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.

[0018] like Figure 1 This application provides a multi-platform data interaction system based on voice interaction, including: Historical voice files of communication with different agents are obtained from the communication system based on the customer ID. These files contain customer ratings of agents and information on whether the quality inspection is qualified. In order to eliminate the impact of communication network instability on quality inspection, it is necessary to calculate the communication network stability of each historical voice file. Factors such as signal strength, packet loss rate, latency, and jitter will change the transmission quality of voice. Information entropy weighting can assign corresponding weights to each factor based on the degree of uncertainty, and more scientifically reflect the overall stability of the network. Using information entropy weighting, these network factors can be combined to calculate the network stability score of each voice file during the call, set a network stability score threshold, and filter out voice files with stability scores greater than the threshold. The steps for calculating the stability score of a communication network using the information entropy weighting method are as follows: Collect network data from historical voice files, including signal strength, packet loss rate, latency, and jitter; For positive indicators, the larger the value, the better the network stability. For example, signal strength can be normalized. For negative metrics, the smaller the value, the better the network stability. For example, packet loss rate, latency, and jitter are inversely normalized. By reverse normalization, it is converted into a value that is "the larger the better", ensuring that all indicators are logically consistent in subsequent calculations; For each index i, calculate its probability in all samples. Calculate the information entropy Entropy measures the degree of fluctuation of an indicator. The higher the entropy, the more dispersed the distribution of the indicator in the sample, and the smaller the impact on stability. Here, n is the number of samples. According to the formula Calculate the weight of each indicator. The larger the entropy value, the greater the fluctuation of the indicator and the lower its contribution to stability. Therefore, the smaller the weight, where m is the number of indicators. According to the formula Calculate a network stability score; the higher the score, the better the network stability. To more effectively analyze voice recordings of varying quality, a rating threshold 'a' was determined through statistical analysis of rating data from a large number of historical recordings. Based on the customer's rating (G) of the agent, recordings with G <= a were classified as negative, indicating customer dissatisfaction with the agent's service; those with G > a were classified as positive, indicating customer satisfaction with the agent's service. This classification method facilitates further in-depth analysis of voice recordings at different rating levels. For example, statistical analysis of the rating data from 800 selected recordings revealed that most ratings were concentrated between 30 and 100. After data fitting and business needs assessment, a rating threshold 'a' was determined to be 70. 'a' can be adjusted according to business requirements. If the customer's rating of the agent is G = 65, which is less than 70, the recording is classified as negative; if G = 80, which is greater than 70, it is classified as positive.

[0019] For the voice files that failed the quality inspection and were rated as poor and passed the quality inspection and were rated as excellent, the voices of the agents were extracted and the basic attributes of these voices were calculated. The basic features include: volume, speaking speed, professionalism, attitude and timbre. The basic attributes that have the greatest impact on customers were then identified. Unstable volume can make customers feel uncomfortable while listening, making it difficult for them to concentrate on the information conveyed by the agent, thus affecting communication effectiveness and customer satisfaction. Quantitative analysis of volume can help determine if the agent has any problems in this regard. Sound pressure level is used to reflect the volume of a sound. A stable sound should be kept within a reasonable sound pressure level. If the sound pressure level frequently exceeds the range, it indicates that the volume of the sound fluctuates. Therefore, the number of time points where the sound pressure level is outside the specified range (num1) and the total number of time points (num2) are counted on the speech timeline. The sound is calculated as sound = num1 / num2. If sound > Num, where Num is a set proportional threshold, it is determined that the volume of the agent's voice is inconsistent, which may affect the listening experience of the customer. A suitable speaking speed helps customers understand the information conveyed by the agent. Too fast may make it difficult for customers to understand, while too slow may make customers lose patience, both of which are detrimental to the communication effect. Count the number of pauses fre, pause duration ti, and total speech duration T in the agent's speech to obtain the actual speech duration (T-sum(ti)). Then combine it with the number of words or phrases in the text to calculate the speaking speed spe=word / (T-sum(ti)). If spe∈[spe1,spe2], it means that the speaking speed is moderate and meets the customer's requirements. The professionalism of the agents is directly related to the customer's trust and satisfaction with the company. The customer follow-up question rate can reflect the accuracy of the agent's answers, while the frequency of professional terms can reflect the agent's mastery of professional knowledge. Combining these two indicators can provide a more comprehensive and accurate assessment of the agent's professionalism. Record the total number of seat answers Ans and the number of customer follow-up questions ans, calculate the customer follow-up rate pro = ans / Ans. Relying solely on the follow-up rate to judge professionalism is not comprehensive enough. It is also necessary to analyze the seat answers in detail after each follow-up, count the number of professional words Np and the total number of words Nt, calculate the professional word frequency PWFi of a single answer as PWFi = Np / Nt, and take the average value of PWFi for all follow-up answers as the comprehensive SPWF of the seat. When pro <= proth, directly determine that the seat is professional; when pro > proth, calculate the weighted score SS = pro * (1 - SPWF). If SS < Sth, it is determined to be professional, otherwise it is unprofessional. Here, proth is the set follow-up rate threshold, and Sth is the set weighted score threshold; A good attitude can make customers feel respected and cared for, enhancing their favorability towards the enterprise. The politeness index and response timeliness are important indicators to measure the seat's attitude. By quantifying these two indicators, the service attitude of the seat can be objectively evaluated; Collect the number of polite words Pos, the number of negative words Neg, the total number of words Total, the number of responses res within a specified time, and the total number of interactions Res. Calculate the text politeness index TPI = (Pos - Neg) / Total, and at the same time calculate the response timeliness RT = res / Res. When TPI >= T1 and RTI >= T2, it is determined that the seat has a good attitude, otherwise the attitude is poor. Here, T1 and T2 are the set thresholds; Different voices will bring different psychological feelings to customers. A pleasant voice may make customers more愉悦 during communication, while a poor voice may affect the customer experience to a certain extent; The human auditory system is sensitive to the sound frequency distribution. The spectral centroid represents the center of the spectral energy distribution. Therefore, the preference for voice can be reflected by calculating the spectral centroid. Obtain the spectral centroid SC. When SC ∈ [SCmin, SCmax], it is determined that the voice is good, otherwise the voice is poor. SCmin and SCmax are the set spectral centroid ranges; Obtain the sound volume, speaking speed, professionalism, attitude, and voice of the voice files that fail the quality inspection and are rated as poor. Record the number of times aa of inconsistent sound volume, inappropriate speaking speed, unqualified professionalism, poor attitude, and poor voice. If aa > [0.8 * AA], it means that the abnormal frequency of this basic attribute in the recorded files that fail the quality inspection and are rated as poor is relatively high, and it is classified as Key Attribute 1. Here, AA is the total number of recorded files that fail the quality inspection and are rated as poor, and [] means rounding; Key Attribute 1 reflects the main factors that lead to the unqualified service quality of the seat. By identifying these basic attributes, the basic attributes that cause customers to be dissatisfied with the seat service can be clearly identified, so that the enterprise can make targeted improvements and trainings; The volume, speed, professionalism, attitude, and timbre of the audio files that pass the quality inspection and are rated as excellent are recorded. The number of times the volume is consistent, the speed is moderate, the professionalism is qualified, the attitude is good, and the timbre is good is recorded as bb. If bb > [0.8*BB], it means that the basic attribute has a high frequency of abnormality in the audio files that pass the quality inspection and are rated as excellent. It is classified as the second key attribute. Here, BB is the total number of audio files that pass the quality inspection and are rated as excellent, and [] means rounding. Key attribute two reflects the key factors for qualified agent service quality. By identifying these basic attributes, we can quickly locate the basic attributes that customers value most in agent service, which is beneficial for further analysis of voice quality inspection below. Different characteristics have varying degrees of impact on customer evaluations. By classifying key attributes into different levels, quality inspection can be conducted in a hierarchical manner based on the importance of basic attributes, thereby improving the efficiency and accuracy of quality inspection. The first and second key attributes are processed, and their intersection is calculated and used as the primary quality inspection attribute. The core purpose of this is to identify common factors that decisively influence customer reviews. Among the many factors affecting customer reviews, some are key factors that play a crucial role regardless of whether the audio file is ultimately rated as acceptable or unacceptable. By finding the intersection, these core factors can be precisely located. The difference between the first and second key features is calculated. The basic attributes in this difference will influence customers to give negative reviews; these are used as the first secondary quality inspection attribute. This is mainly to clarify which... The factors that primarily lead to negative customer reviews, but are not the sole determinants of positive reviews, are the key factors that need to be focused on during voice quality inspection. This calculation method can highlight these critical negative factors. The difference between key feature two and key feature one is calculated; the fundamental attributes within this difference will influence positive customer reviews, and these are treated as secondary quality inspection attributes two. The significance of this is to uncover those factors that primarily determine positive customer reviews, but are not the sole determinants of negative reviews. In voice quality inspection, in addition to focusing on factors that may lead to negative reviews, it is also necessary to identify which factors can improve customer satisfaction and achieve a higher level of service. For example, suppose the first key attribute is {voice volume, speaking speed, professionalism}, and the second key attribute is {speaking speed, attitude, tone}. The primary quality inspection attribute is {speaking speed}. Taking "speaking speed" as an example, an appropriate speaking speed is crucial for ensuring smooth communication. It affects both the customer's reception of information and reflects the agent's professionalism. Using it as the primary quality inspection attribute allows for a quick assessment of whether the agent meets the core communication skills standards at the initial stage of the quality inspection process, providing crucial evidence for subsequent quality inspection steps and improving efficiency and accuracy. The secondary quality inspection attribute is {voice volume, professionalism}. Taking "voice volume and professionalism" as examples, an inappropriate voice volume or unprofessional answers may cause customers to... Discomfort directly impacts customer experience, leading to negative reviews. When primary quality inspection attributes fail or have issues, targeted checks on these factors that easily trigger negative reviews help identify and resolve key issues in agent service in a timely manner, providing a clear direction for improving service quality. The second secondary quality inspection attribute is {attitude and tone of voice}. Taking "tone of voice and attitude" as an example, a good tone of voice and a positive attitude can enhance the customer's communication experience and increase the likelihood of positive reviews. If the primary quality inspection attributes pass, further evaluation of agent service performance in improving customer satisfaction ensures that service quality not only meets basic requirements but also reaches higher standards, comprehensively improving the customer experience. Distinguishing between controllable and uncontrollable attributes allows companies to allocate resources more rationally during quality inspection and training. For controllable attributes, the performance of agents can be improved through training and management. For uncontrollable attributes, appropriate consideration can be given during quality inspection to avoid unfair evaluations of agents. Further analysis of the identified key attributes one and key attributes two revealed that basic attributes such as volume, speaking speed, professionalism, and attitude can be controlled by agents through training and self-adjustment, and are therefore classified as controllable quality inspection attributes. However, timbre is greatly affected by innate factors and is difficult to control, so it is classified as an uncontrollable quality inspection attribute. Controllable and uncontrollable attributes are recorded separately and treated differently during quality inspection and agent training.

[0020] Based on the customer ID, locate the audio file to be inspected. If the file is rated as excellent, first check the primary quality inspection attribute. If it is within the specified threshold, proceed to verify the secondary quality inspection attribute. If both are within the specified threshold, the quality inspection is qualified. If the primary quality inspection attribute is not within the specified threshold, but it is found to contain an uncontrollable quality inspection attribute and only that uncontrollable quality inspection attribute is not within the specified threshold, then proceed to the secondary quality inspection attribute. If it passes, the quality inspection is qualified; otherwise, the quality inspection is unqualified. If the primary quality inspection attribute is not within the specified threshold and does not contain an uncontrollable quality inspection attribute, the quality inspection is directly deemed unqualified. For excellent-rated voice files, the primary quality inspection attribute is the factor that determines the customer's evaluation. If it is within the specified threshold, it indicates that the agent performs well in key aspects. At this time, verifying the secondary quality inspection attribute two can further ensure the comprehensiveness of service quality. When the primary quality inspection attribute is not within the specified threshold and contains uncontrollable factors, to avoid misjudgment caused by uncontrollable factors, it is necessary to re-judge whether only this uncontrollable quality inspection attribute exceeds the specified threshold in the primary quality inspection attribute. If so, give the opportunity to verify the secondary quality inspection feature two; if there are other basic attributes in the primary quality inspection attribute exceeding the specified threshold in addition to the uncontrollable quality inspection attribute, it means that the agent has problems in key aspects and should be determined as unqualified and receive targeted training; For example, the file to be quality inspected is evaluated as excellent. The primary quality inspection attribute is the speech rate, and the specified threshold is 120-180 words per minute. After detection, the agent's speech rate is 150 words per minute, which is within the specified threshold. The secondary quality inspection feature two is attitude and tone color. The text politeness index TPI = 0.3 in terms of attitude, the response timeliness RTI = 0.9, and the spectral centroid SC of the tone color = 300Hz, all of which meet the requirements and the verification passes, so the quality inspection of this file is qualified; if the primary quality inspection attribute includes the speech rate and tone color, the tone color is an uncontrollable quality inspection attribute, and the secondary quality inspection attribute two is verified and passed, then the quality inspection of this file is also qualified. If the speech rate is not within the specified threshold, the quality inspection is unqualified; If the file grade is poor, first check the primary quality inspection attribute. If it is not within the specified range, turn to the secondary quality inspection attribute one. If there is a basic attribute not within the specified threshold, the quality inspection is unqualified; if there is an uncontrollable quality inspection attribute in the primary quality inspection attribute and only this basic attribute is not within the specified threshold, then turn to the secondary quality inspection attribute two. If all of them are within the specified threshold range, it means that the agent is affected by uncontrollable factors rather than its own professional problems, and the quality inspection can be determined as qualified; if the primary quality inspection is within the specified range, turn to the secondary quality inspection attribute two. If all of them are within the specified threshold range, the quality inspection is qualified, otherwise the quality inspection is unqualified; For poor-rated voice files, when the primary quality inspection attribute is not within the specified range, it is necessary to further verify whether there is an uncontrollable quality inspection attribute. If there is an uncontrollable quality inspection attribute and only this basic attribute is not within the specified threshold range, and at the same time all the secondary quality inspection attributes two are within the specified threshold range, the determination of being qualified takes into account the impact of uncontrollable factors on service quality; For example, the file to be quality inspected is evaluated as poor. The primary quality inspection attribute is the speech rate. After detection, the speech rate is not within the specified range. Turn to the secondary quality inspection attribute one for verification. If the sound volume is also not within the specified threshold, the quality inspection is unqualified; if the primary quality inspection attribute includes the tone color, the tone color is an uncontrollable quality inspection attribute, and the spectral centroid SC of the tone color is not within the specified range. If all the basic attributes of the secondary quality inspection attribute two are within the threshold range, the quality inspection is determined as qualified, otherwise, the quality inspection is unqualified.

[0021] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0022] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A multi-platform data interaction system based on voice interaction, characterized in that, include: Based on the customer ID, retrieve historical voice files of communication with different agents. Calculate the communication network stability score S using the information entropy weighting method. Filter out voice files where S>Sth. Obtain primary quality inspection attributes, secondary quality inspection attribute one, secondary quality inspection attribute two, controllable quality inspection attributes, and uncontrollable quality inspection attributes based on key attribute one and key attribute two. The historical voice files contain customer ratings and quality inspection information. Based on the customer ID, locate the voice file to be inspected. If G>a, check the primary quality inspection attribute. If it meets the requirement, proceed to the secondary quality inspection attribute. If it meets the requirement, the quality inspection is qualified. If the primary quality inspection attribute does not meet the requirement, but it contains an uncontrollable quality inspection attribute and only that attribute is not within the specified threshold, proceed to the secondary quality inspection attribute. If it meets the requirement, the quality inspection is qualified; otherwise, the quality inspection is unqualified. If the primary quality inspection attribute does not meet the requirement except for the uncontrollable quality inspection attribute, the quality inspection is directly judged as unqualified. Here, G is the agent score and a is the scoring threshold. If G <= a, check the primary quality inspection attribute. If it does not meet the requirements, proceed to secondary quality inspection attribute one. If any basic attribute is outside the specified threshold, the quality inspection fails. If there is an uncontrollable quality inspection attribute among the primary quality inspection attributes and only that basic attribute is outside the specified threshold, proceed to secondary quality inspection attribute two. If both are within the specified threshold range, the quality inspection passes. If the primary quality inspection attribute meets the requirements, proceed to secondary quality inspection attribute two. If it meets the requirements, the quality inspection passes; otherwise, the quality inspection fails.

2. The multi-platform data interaction system based on voice interaction according to claim 1, characterized in that, The specific steps for calculating the stability score S of a communication network using the information entropy weighting method are as follows: Collect network data from historical voice files, specifically signal strength, packet loss rate, latency, and jitter. According to the formula The signal strength is normalized according to the formula. Normalize the packet loss rate, latency, and jitter. For each index i, calculate its probability in all samples. Calculate the information entropy Where n is the number of samples; According to the formula Calculate the weight of each indicator, where m is the number of indicators; According to the formula Calculate the network stability score.

3. The multi-platform data interaction system based on voice interaction according to claim 1, characterized in that, The basic attributes of voice files with G<=a and failing quality inspection and those with G>a and passing quality inspection are calculated. The basic attributes include volume, speech rate, professionalism, attitude, and timbre.

4. The multi-platform data interaction system based on voice interaction according to claim 3, characterized in that, The specific steps for determining whether voice volume, speaking speed, professionalism, attitude, and timbre are within the specified thresholds are as follows: The volume of the voice is determined by counting the number of time points (num1) where the sound pressure level is outside the specified range and the total number of time points (num2) on the speech time axis. The sound is calculated as sound = num1 / num2. If sound ∈ [Num1, Num2], the volume of the voices of the seats is determined to be consistent. Here, Num1 and Num2 are the set ratio thresholds. Speech rate is calculated by counting the number of pauses (fre), pause duration (ti), and total speech duration (T) in the voice of the agent, obtaining the actual speech duration (T-sum(ti)), and then calculating the speech rate (spe=word / (T-sum(ti)). If spe∈[spe1,spe2], it indicates that the speech rate is moderate, where word is the number of characters or words in the text content, and spe1 and spe2 are the set speech rate thresholds. Professionalism: Record the total number of answers Ans by the agent and the number of customer follow-up questions ans, calculate the customer follow-up rate pro = ans / Ans, count the number of professional words Np and the total number of words Nt, calculate the professional word frequency PWFi for each answer as PWFi = Np / Nt, and take the average value of PWFi for all follow-up answers as the comprehensive SPWF of the agent. When pro <= proth, directly determine that the agent is professional. When pro > proth, calculate SS = pro * (1 - SPWF). If SS < Sth, then it is determined to be professional; otherwise, it is unprofessional. Here, proth is the set follow-up rate threshold, and Sth is the set weighted score threshold; Attitude: Collect the number of polite words Pos, the number of negative words Neg, the total number of words Total, the number of responses res within a specified time, and the total number of interactions Res. Calculate the text politeness index TPI = (Pos - Neg) / Total, and at the same time calculate the response timeliness RT = res / Res. When TPI >= T1 and RTI >= T2, it is determined that the agent has a good attitude; otherwise, the attitude is poor. Here, T1 and T2 are the set thresholds; Voice quality: Obtain the spectral centroid SC. When SC ∈ [SCmin, SCmax], it is determined that the voice quality is good; otherwise, the voice quality is poor. SCmin and SCmax are the set spectral centroid ranges.

5. The multi-platform data interaction system based on voice interaction according to claim 1, characterized in that, Record the number of times aa that each basic attribute of the voice files with G <= a and unqualified quality inspection is not within the threshold range. If aa > [0.8 * AA], then classify it as key attribute one. Record the number of times bb that each basic attribute of the voice files with G > a and qualified quality inspection is within the threshold range. If bb > [0.8 * BB], then classify it as key attribute two. Here, AA is the total number of recording files that are unqualified and rated as poor, and BB is the total number of recording files that are qualified and rated as excellent. [] represents the rounding symbol.

6. The multi-platform data interaction system based on voice interaction according to claim 1, characterized in that, Find the intersection of key attribute one and key attribute two, and use it as the primary quality inspection attribute. Find the difference set between key attribute one and key attribute two, and use it as the secondary quality inspection attribute one. Find the difference set between key attribute two and key attribute one, and use it as the secondary quality inspection attribute two.

7. The multi-platform data interaction system based on voice interaction according to claim 1, characterized in that, The controllable quality inspection attributes include sound volume, speaking speed, professionalism, and attitude. The uncontrollable quality inspection attribute includes voice quality.

8. The multi-platform data interaction system based on voice interaction according to claim 1, characterized in that, When the primary quality inspection attribute is an empty set, when G > a, turn to the secondary quality inspection attribute two. If all meet the threshold range, the quality inspection is qualified; otherwise, the quality inspection is unqualified. When G <= a, turn to the secondary quality inspection attribute one. If there is any that is not within the threshold range, the quality inspection is unqualified. If the one not within the threshold range is an uncontrollable quality inspection attribute and other basic attributes are all controllable attributes, then turn to the secondary quality inspection attribute one. If all meet the threshold range, the quality inspection is qualified.

Images