Outbound system interaction method based on voice recognition
By constructing an accent judgment model and using speech recognition technology, the system can accurately identify customer emotions and calculate anger levels, dynamically adjust compensation strategies, and solve the problem that traditional outbound calling systems cannot identify the degree of customer anger, thereby improving customer experience and marketing effectiveness.
Patent Information
- Application Number
- CN202511476134.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-16
- Publication Date
- 2026-01-23
AI Technical Summary
Traditional outbound calling systems cannot accurately identify customer emotions or quantify the degree of customer anger, resulting in poor customer experience, low marketing success rate, and a high likelihood of complaints.
An accent detection model is built based on a convolutional neural network. By combining a speech recognition database and a machine learning model, customer emotion data and speech features are analyzed to calculate the emotion perception index and anger level coefficient. Real-time compensation is then provided in the form of coupons.
It enables precise identification and quantification of customer emotions, dynamically adjusts compensation strategies, improves customer satisfaction, reduces complaint risks, and enhances the practicality and competitiveness of marketing and customer service.
Smart Images
Figure CN121397147A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of outbound call system interaction, and particularly relates to an outbound call system interaction method based on voice recognition. BACKGROUND
[0002] The outbound call system is an automatic telephone dialing system based on CTI (Computer Telephone Integration) technology, which dials user's telephone in batches through a computer and plays a preset voice, and is mainly used to improve the efficiency of telephone marketing, customer service and the like. Its core functions include automatic dialing, call recording management, customer information screening and intelligent call distribution. The outbound call system originated from the computer telephone integration application in the late 20th century, and was mainly used for simple voice playing in the early stage. With the progress of technology, intelligent routing, multi-channel interaction and other functions are gradually integrated, forming three mainstream modes of predictive, preview and reservation: the predictive mode dynamically adjusts the dialing rhythm through an algorithm; the preview mode needs manual intervention for call confirmation; and the reservation mode realizes directional service based on customer active reservation.
[0003] However, the traditional outbound call system lacks refinement in customer emotion processing, lacks identification of the emotion category of the customer, cannot quantify the intensity or anger level of the customer emotion, and cannot dynamically adjust the compensation strategy according to the emotion intensity, thereby leading to poor customer experience, low marketing success rate, and even customer complaints.
[0004] In view of the above technical defects, the present application provides a solution. SUMMARY
[0005] The present application aims to solve the problem that the traditional outbound call system lacks refinement in customer emotion processing, can only identify simple emotion categories, cannot quantify the intensity or anger level of the customer emotion, and cannot dynamically adjust the compensation strategy according to the emotion intensity, thereby leading to poor customer experience, low marketing success rate, and even customer complaints.
[0006] In order to achieve the above purpose, the present application adopts the following technical scheme: an outbound call system interaction method based on voice recognition, comprising the following steps:
[0007] Step one, an accent judgment model is constructed based on a convolutional neural network to judge the region to which the customer belongs, and a regional emotion adjustment coefficient is preset according to different regions;
[0008] Step two, through a voice recognition database, the voice samples of the call between the outbound call system and the customer are denoised, and emotion data, voice feature data and flat data of the voice samples are collected;
[0009] Step 3: Analyze and calculate the emotion perception index by using the preset regional emotion coefficient and emotion data. At the same time, classify customer emotions into satisfied emotions and angry emotions, and analyze the anger level coefficient of angry customers.
[0010] Step 4: By acquiring voice feature data, analyze and calculate the customer's anger level coefficient, which reflects the severity of the customer's anger.
[0011] Step 5: By obtaining calming data and combining it with the anger level coefficient, calculate the emotional discount for angry customers, inform them of the discount coupon, and issue it to them.
[0012] Furthermore, the process of constructing the accent detection model is as follows:
[0013] S11. Collect speech sample data of accents from multiple different regions as a dataset, and randomly divide the dataset into a training set and a test set. Label the corresponding regions on the accents in the training set as labels.
[0014] S12. Construct an accent judgment model based on a convolutional neural network, train the accent judgment model using a training set, and test the accent judgment model using a test set to obtain a qualified accent judgment model.
[0015] S13. Obtain the customer's real-time accent data through the speech recognition database, and input the customer's real-time accent data into the qualified accent judgment model to determine the customer's location.
[0016] S14. By analyzing historical outbound call data, a machine learning model is used for training. The feature weights learned by the model are the regional sentiment coefficients, which are used to reflect the degree of influence of different accents on sentiment judgment.
[0017] Furthermore, the emotion data includes the text emotion score of the customer's call text, the volume decibels of the customer's call, and the pitch frequency data of the customer's call; the voice feature data includes the fundamental frequency of the customer's call voice, the speech rate during the customer's call, and the amplitude of voice energy fluctuations during the customer's call; and the calming data includes the number of times the customer's call voice was interrupted and the preset basic discount amount data.
[0018] Furthermore, the calculation process for the customer's emotional perception index is as follows:
[0019] S21. Obtain the text sentiment score of the customer's call text, the volume decibel of the customer's call, and the tone frequency data of the customer's call, and analyze and calculate them in combination with the preset regional sentiment coefficient.
[0020] S22. Calculate the customer's emotional perception index S according to the following formula:
[0021]
[0022] Where f is the text sentiment score of the customer's call text, time is the total duration of the call interaction between the outbound calling system and the customer, and L i Let L be the volume of the customer's call in decibels at time i. e F is the preset standard volume decibel for calls. i Let F be the tone frequency of the customer's call at time i. e δ is the preset standard tone frequency for calls, a is the preset weighting coefficient for volume decibels, b is the preset weighting coefficient for tone frequency, and δ is the preset regional mood adjustment coefficient based on the customer's region.
[0023] S23. Obtain the preset emotion perception index S th A comparative analysis was conducted with the emotion perception index S. When S > S th If the customer's emotional satisfaction level is low during the outbound call conversation, the customer's emotions will be classified as anger, and the anger level coefficient of the angry customer will be analyzed.
[0024] S24, when S≤S th This indicates that the customer's emotional satisfaction level is high during the outbound call conversation, and the customer's emotions are classified as satisfied emotions.
[0025] Furthermore, the calculation process for the customer's anger level coefficient is as follows:
[0026] S31. Obtain the base frequency of the customer's voice during the call, the customer's speech rate during the call, and the amplitude of voice energy fluctuations during the customer's call, and perform analysis and calculation.
[0027] S32. Calculate the customer's anger level coefficient X according to the following formula:
[0028]
[0029] Where Δp is the rate of change of the fundamental frequency of the customer's voice during the call, p max p is the preset maximum base frequency for customer call audio. min Here, v is the preset minimum fundamental frequency for the customer's voice during the call, and v is the customer's speech rate during the call. max v is the preset maximum speaking speed during a customer call. min D represents the preset minimum speech rate during a customer call, and D represents the amplitude of speech energy fluctuation during the customer call. eThe preset standard fluctuation range of voice energy is α, the preset weighting coefficient of the call voice fundamental frequency is β, the preset weighting coefficient of the speech rate is γ, and the preset weighting coefficient of the voice energy fluctuation range is γ, and α+β+γ=1. The customer's anger level coefficient is used to reflect the severity of the customer's anger. The larger the value of the customer's anger level coefficient, the more severe the customer's anger is.
[0030] Furthermore, the calculation process for the emotional discount for angry customers is as follows:
[0031] S41. Obtain data on the number of times a customer's voice call is interrupted and the preset basic discount amount, and analyze and calculate it in conjunction with the customer's anger level coefficient;
[0032] S42. Calculate the emotional discount H for angry customers according to the following formula:
[0033]
[0034] Where B is the preset base discount amount, and T is the number of times the customer's voice call is interrupted within a preset unit of time. max The preset maximum number of customer call interruptions, k is the preset customer emotion adjustment factor, and X is the customer's anger level coefficient.
[0035] S43. The calculated emotional discount is presented to the customer in the form of a coupon via voice and text message, and automatically issued to the customer's account through the outbound call system interface to calm the customer's anger.
[0036] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:
[0037] This speech recognition-based outbound calling system interaction method combines regional emotion coefficients and emotion data, and uses a formula to scientifically calculate an emotion perception index. This allows for the objective differentiation of customer emotions into satisfaction or anger, with refined identification of anger. Furthermore, by calculating an anger level coefficient using speech feature data, the severity of customer anger is quantified, providing a precise basis for subsequent compensation strategies. Moreover, based on calming data and the anger level coefficient, emotional incentives are dynamically calculated and automatically distributed as coupons via voice and text messages. This achieves real-time response and compensation for customer emotions, effectively calming customer anger, significantly improving customer satisfaction, reducing complaint risks, and ultimately enhancing the practicality and competitiveness of the outbound calling system in marketing and customer service scenarios. Attached Figure Description
[0038] Figure 1 A schematic diagram of the method flow of the present invention is shown. Detailed Implementation
[0039] 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.
[0040] Example:
[0041] like Figure 1 As shown, an interactive method for an outbound calling system based on speech recognition is firstly constructed based on a convolutional neural network to determine the customer's region, and then a regional sentiment adjustment coefficient is preset according to different regions.
[0042] The process of constructing the accent detection model is as follows:
[0043] S11. Collect speech sample data of accents from multiple different regions as a dataset, and randomly divide the dataset into a training set and a test set. Label the corresponding regions on the accents in the training set as labels.
[0044] S12. Construct an accent judgment model based on a convolutional neural network, train the accent judgment model using a training set, and test the accent judgment model using a test set to obtain a qualified accent judgment model.
[0045] S13. Obtain the customer's real-time accent data through the speech recognition database, and input the customer's real-time accent data into the qualified accent judgment model to determine the customer's location.
[0046] S14. By analyzing historical outbound call data (including accent labels and emotion perception index results), a machine learning model (such as logistic regression) is used for training. The feature weights learned by the model are the regional emotion coefficients, which are used to reflect the degree of influence of different accents on emotion judgment.
[0047] Then, noise reduction processing is performed on the voice samples of the outbound call system's calls with customers through the speech recognition database, and emotion data, voice feature data and calming data of the voice samples are collected.
[0048] Sentiment data includes the text sentiment score of the customer's call text, the volume decibels of the customer's call, and the frequency of the customer's tone of voice. The text sentiment score is obtained through a sentiment classifier and a sentiment dictionary. The sentiment classifier is a trained machine learning or deep learning model that analyzes and classifies input text, determines the sentiment tendency expressed in the text, and quantifies it as a score. For example, neural network-based sentiment analysis models, such as recurrent neural networks (RNNs), long short-term memory networks (LSTMs), or Transformer architecture models (such as BERT), can be used. These models are trained on large amounts of text data labeled with sentiment tags, learning language patterns and sentiment expressions in the text, thus enabling them to classify and score new text based on sentiment. The sentiment dictionary contains a large number of words and their corresponding sentiment polarities and intensities. When extracting the text sentiment score using the sentiment dictionary, each word in the text is matched against the sentiment dictionary, and the sentiment score of the entire text is calculated based on the sentiment polarity (positive, negative) and intensity of the matched words. For example, if a text contains multiple positive sentiment words with high sentiment intensity, the text's sentiment score will be relatively high; conversely, if it contains many negative sentiment words, the sentiment score will be low. Voice feature data includes the fundamental frequency of the customer's voice during the call, the customer's speech rate during the call, and the amplitude of voice energy fluctuations during the call. The smoothing data includes the number of times the customer's voice was interrupted and the preset basic discount amount.
[0049] Then, by using a preset regional emotion coefficient and combining it with emotion data, an emotion perception index is calculated. At the same time, customer emotions are divided into satisfied emotions and angry emotions, and the anger level coefficient of angry customers is analyzed.
[0050] The calculation process for the customer's emotional perception index is as follows:
[0051] S21. Obtain the text sentiment score of the customer's call text, the volume decibel of the customer's call, and the tone frequency data of the customer's call, and analyze and calculate them in combination with the preset regional sentiment coefficient.
[0052] S22. Calculate the customer's emotional perception index S according to the following formula:
[0053]
[0054] Where f is the text sentiment score of the customer's call text, time is the total duration of the call interaction between the outbound calling system and the customer, and L i Let L be the volume of the customer's call in decibels at time i. e F is the preset standard volume decibel for calls. iLet F be the tone frequency of the customer's call at time i. e δ is the preset standard tone frequency for calls, a is the preset weighting coefficient for volume decibels, b is the preset weighting coefficient for tone frequency, and δ is the preset regional mood adjustment coefficient based on the customer's region.
[0055] S23. Obtain the preset emotion perception index S th A comparative analysis was conducted with the emotion perception index S. When S > S th If the customer's emotional satisfaction level is low during the outbound call conversation, the customer's emotions will be classified as anger, and the anger level coefficient of the angry customer will be analyzed.
[0056] S24, when S≤S th If the customer's emotional satisfaction level is high during the outbound call, the outbound call system will not adjust the current script or tone of voice and will continue to communicate with the customer.
[0057] Then, by acquiring voice feature data, the anger level coefficient of the customer is analyzed and calculated. The anger level coefficient is used to reflect the severity of the customer's anger.
[0058] The calculation process for the customer's anger level coefficient is as follows:
[0059] S31. Obtain the base frequency of the customer's voice during the call, the customer's speech rate during the call, and the amplitude of voice energy fluctuations during the customer's call, and perform analysis and calculation.
[0060] S32. Calculate the customer's anger level coefficient X according to the following formula:
[0061]
[0062] Where Δp is the rate of change of the fundamental frequency of the customer's voice during the call, p max The preset maximum fundamental frequency of the customer's voice during a call (the fundamental frequency is the basic frequency generated by the vibration of the vocal cords, and its periodic fluctuations reflect the tension and stability of the vocal cords), p min Here, v is the preset minimum fundamental frequency for the customer's voice during the call, and v is the customer's speech rate during the call. max v is the preset maximum speaking speed during a customer call. min D represents the preset minimum speech rate during a customer call, and D represents the amplitude of speech energy fluctuation during the customer call. eThe preset standard fluctuation range of voice energy is α, the preset weighting coefficient of the call voice fundamental frequency is β, the preset weighting coefficient of the speech rate is γ, and the preset weighting coefficient of the voice energy fluctuation range is γ, and α+β+γ=1. The customer's anger level coefficient is used to reflect the severity of the customer's anger. The larger the value of the customer's anger level coefficient, the more severe the customer's anger is.
[0063] Finally, by acquiring calming data and combining it with the anger level coefficient, the emotional discount for angry customers is calculated, and then distributed to customers in the form of coupons.
[0064] The calculation process for the emotional discount for angry customers is as follows:
[0065] S41. Obtain data on the number of times a customer's voice call is interrupted and the preset basic discount amount, and analyze and calculate it in conjunction with the customer's anger level coefficient;
[0066] S42. Calculate the emotional discount H for angry customers according to the following formula:
[0067]
[0068] Where B is the preset base discount amount, and T is the number of times the customer's call is interrupted within a preset unit of time (e.g., a 10-second window). max The preset maximum number of customer call interruptions, k is the preset customer emotion modulation factor (determined through a large amount of experimental data), and X is the customer's anger level coefficient.
[0069] S43. The calculated emotional discount is presented to the customer in the form of a coupon via voice and text message, and automatically issued to the customer's account through the outbound call system interface to calm the customer's anger.
[0070] This invention combines regional emotion coefficients and emotion data, and uses a formula to scientifically calculate an emotion perception index, which can objectively distinguish customer emotions into satisfaction or anger, and perform refined identification of anger. Furthermore, by calculating an anger level coefficient using voice feature data, the severity of customer anger is quantified, providing a precise basis for subsequent compensation strategies. Moreover, based on calming data and the anger level coefficient, emotional discounts are dynamically calculated and automatically distributed in the form of coupons via voice and text message, achieving real-time response and compensation for customer emotions, effectively calming customer anger, significantly improving customer satisfaction, reducing complaint risks, and thus enhancing the practicality and competitiveness of the outbound call system in marketing and customer service scenarios.
[0071] The size of the interval and threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value.
[0072] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.
[0073] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for interaction in an outbound calling system based on speech recognition, characterized in that, Includes the following steps: Step 1: Construct an accent judgment model based on a convolutional neural network to determine the customer's region and preset a regional sentiment adjustment coefficient according to different regions; Step 2: Using the speech recognition database, noise reduction is performed on the voice samples of the outbound call system's conversations with customers, and emotion data, speech feature data, and calming data of the voice samples are collected. Step 3: Analyze and calculate the emotion perception index by using the preset regional emotion coefficient and emotion data. At the same time, classify customer emotions into satisfied emotions and angry emotions, and analyze the anger level coefficient of angry customers. Step 4: By acquiring voice feature data, analyze and calculate the customer's anger level coefficient, which reflects the severity of the customer's anger. Step 5: By obtaining calming data and combining it with the anger level coefficient, calculate the emotional discount for angry customers, inform them of the discount coupon, and issue it to them.
2. The interaction method for an outbound calling system based on speech recognition according to claim 1, characterized in that, The process of constructing the accent detection model is as follows: S11. Collect speech sample data of accents from multiple different regions as a dataset, and randomly divide the dataset into a training set and a test set. Label the corresponding regions on the accents in the training set as labels. S12. Construct an accent judgment model based on a convolutional neural network, train the accent judgment model using a training set, and test the accent judgment model using a test set to obtain a qualified accent judgment model. S13. Obtain the customer's real-time accent data through the speech recognition database, and input the customer's real-time accent data into the qualified accent judgment model to determine the customer's location. S14. By analyzing historical outbound call data, a machine learning model is used for training. The feature weights learned by the model are the regional sentiment coefficients, which are used to reflect the degree of influence of different accents on sentiment judgment.
3. The interaction method for an outbound calling system based on speech recognition according to claim 1, characterized in that, The emotion data includes the text emotion score of the customer's call text, the volume decibels of the customer's call, and the pitch frequency data of the customer's call. The voice feature data includes the base frequency of the customer's call voice, the speech rate during the customer's call, and the amplitude of voice energy fluctuations during the customer's call. The calming data includes the number of times the customer's call voice was interrupted and the preset basic discount amount data.
4. The interaction method for an outbound calling system based on speech recognition according to claim 1, characterized in that, The calculation process for the customer's emotional perception index is as follows: S21. Obtain the text sentiment score of the customer's call text, the volume decibel of the customer's call, and the tone frequency data of the customer's call, and analyze and calculate them in combination with the preset regional sentiment coefficient. S22. Calculate the customer's emotional perception index S according to the following formula: Where f is the text sentiment score of the customer's call text, time is the total duration of the call interaction between the outbound calling system and the customer, and L i Let L be the volume of the customer's call in decibels at time i. e F is the preset standard volume decibel for calls. i Let F be the tone frequency of the customer's call at time i. e δ is the preset standard tone frequency for calls, a is the preset weighting coefficient for volume decibels, b is the preset weighting coefficient for tone frequency, and δ is the preset regional mood adjustment coefficient based on the customer's region. S23. Obtain the preset emotion perception index S th A comparative analysis was conducted with the emotion perception index S. When S > S th If the customer's emotional satisfaction level is low during the outbound call conversation, the customer's emotions will be classified as anger, and the anger level coefficient of the angry customer will be analyzed. S24, when S≤S th This indicates that the customer's emotional satisfaction level is high during the outbound call conversation, and the customer's emotions are classified as satisfied emotions.
5. The interaction method for an outbound calling system based on speech recognition according to claim 1, characterized in that, The calculation process for the customer's anger level coefficient is as follows: S31. Obtain the base frequency of the customer's voice during the call, the customer's speech rate during the call, and the amplitude of voice energy fluctuations during the customer's call, and perform analysis and calculation. S32. Calculate the customer's anger level coefficient X according to the following formula: Where Δp is the rate of change of the fundamental frequency of the customer's voice during the call, p max p is the preset maximum base frequency for customer call audio. min Here, v is the preset minimum fundamental frequency for the customer's voice during the call, and v is the customer's speech rate during the call. max v is the preset maximum speaking speed during a customer call. min D represents the preset minimum speech rate during a customer call, and D represents the amplitude of speech energy fluctuation during the customer call. e The preset standard fluctuation range of voice energy is α, the preset weighting coefficient of the call voice fundamental frequency is β, the preset weighting coefficient of the speech rate is γ, and the preset weighting coefficient of the voice energy fluctuation range is γ, and α+β+γ=1. The customer's anger level coefficient is used to reflect the severity of the customer's anger. The larger the value of the customer's anger level coefficient, the more severe the customer's anger is.
6. The interaction method for an outbound calling system based on speech recognition according to claim 1, characterized in that, The calculation process for the emotional discount for angry customers is as follows: S41. Obtain data on the number of times a customer's voice call is interrupted and the preset basic discount amount, and analyze and calculate it in conjunction with the customer's anger level coefficient; S42. Calculate the emotional discount H for angry customers according to the following formula: Where B is the preset base discount amount, and T is the number of times the customer's voice call is interrupted within a preset unit of time. max The preset maximum number of customer call interruptions, k is the preset customer emotion adjustment factor, and X is the customer's anger level coefficient. S43. The calculated emotional discount is presented to the customer in the form of a coupon via voice and text message, and automatically issued to the customer's account through the outbound call system interface to calm the customer's anger.