Dialogue management method and system of AI intelligent outbound system

Abstract

The invention discloses a dialogue management method and system of an AI intelligent outbound system, and relates to the technical field of intelligent outbound, and the technical key points of the scheme are as follows: carrying out fast Fourier transform on a voice signal to obtain a frequency spectrum, approximating the frequency response of a filter by using a Gaussian function, carrying out point multiplication operation, and summing to obtain the energy of the filter; the method comprises the following steps: calculating a difference value between energy logarithms of adjacent filters, carrying out recursive calculation to obtain a multi-order difference coefficient, calculating to obtain a difference coefficient fluctuation index of each order of difference coefficient, taking an order of a minimum value of the difference coefficient fluctuation index as an order of the multi-order difference coefficient, and calculating a difference value between voice peak intensity and noise intensity. The method comprises the following steps of: obtaining a voice quality evaluation value through a standard deviation of a multi-order difference coefficient, combining the definition index and a signal-to-noise ratio, and performing calculation to obtain the voice quality evaluation value, and when the voice quality evaluation value is smaller than a quality threshold value, switching manual outbound, thereby realizing accurate recognition of a low-quality voice signal.

Description

Technical Field

[0001] The present invention relates to the field of intelligent outbound calling technology, and specifically to a dialogue management method and system for an AI intelligent outbound calling system. Background Art

[0002] With the rapid development of artificial intelligence technology, AI intelligent outbound call system has become an important tool for enterprises to communicate with customers efficiently. Traditional outbound call systems mainly rely on manual operations, and have problems such as low efficiency and high cost. The AI ​​intelligent outbound call system can use natural language processing, speech recognition, machine learning and other advanced technologies to realize automated calls, conversations and data management, greatly improving the company's customer service efficiency and marketing effectiveness.

[0003] In a Chinese invention application with application publication number CN111104502A, a dialogue management method, system, electronic device and storage medium for an outbound call system are disclosed. The dialogue management method includes: obtaining semantic information; updating the current dialogue state according to the semantic information; inputting the dialogue state into a pre-stored state machine for state jump; the outbound call system outputs questions and answers according to the state after the state machine jumps, and also uses reinforcement learning to improve the generalization performance of the dialogue system.

[0004] In a Chinese invention application with application publication number CN115048500A, a method and system for intelligent selection of outbound call dialogue management based on machine learning is disclosed, the method comprising: establishing a business classification model according to user characteristics, pre-training the business classification model using a machine learning algorithm, and generating a business classification result; selecting different dialogue management modules according to the business classification result and sending them to an AI outbound call device; using the AI ​​outbound call device to execute different outbound call dialogue strategies according to the selected dialogue management module and run the corresponding outbound call task.

[0005] In combination with the above invention applications, the prior art has the following deficiencies:

[0006] 1. Existing technologies often only perform simple calculations and comparisons on the difference coefficients of a single order, lacking in-depth analysis and processing. This results in the inability of the intelligent outbound call system to accurately reflect the characteristic changes of voice signals, and to effectively adapt to the voice characteristics and habits of different users. In particular, when processing complex or diverse voice inputs, the intelligent outbound call system may have recognition errors or comprehension deviations.

[0007] 2. The intelligent outbound call system may only focus on certain aspects of the voice signal, such as signal-to-noise ratio or clarity, but ignore other important factors, such as voice fluency, naturalness or emotional expression, making it difficult to accurately judge whether the quality of the voice signal meets the requirements. When encountering low-quality voice signals, the intelligent outbound call system may not be able to make a correct response in time, resulting in misunderstandings or unrecognizable problems. Summary of the invention

[0008] 1. Technical issues to be solved

[0009] In view of the deficiencies in the prior art, the present invention provides a dialogue management method and system for an AI intelligent outbound call system, which obtains a spectrum by fast Fourier transforming a voice signal, approximates the frequency response of a filter using a Gaussian function, performs a point multiplication operation, and sums up to obtain the energy of the filter, obtains a first-order difference coefficient by calculating the difference between the logarithms of the energies of adjacent filters, performs a recursive calculation to obtain multi-order difference coefficients, calculates a difference coefficient fluctuation index for each order of difference coefficient, takes the order at which the minimum value of the difference coefficient fluctuation index is located as the order of the multi-order difference coefficient, calculates the difference between the speech peak intensity and the noise intensity, and sums up to obtain a clarity index, calculates the standard deviation of the multi-order difference coefficients through the multi-order difference coefficients, combines the clarity index and the signal-to-noise ratio to obtain a voice quality evaluation value, and when the voice quality evaluation value is less than the quality threshold, switches to manual outbound calling, thereby solving the problems mentioned in the background technology.

[0010] (II) Technical solution

[0011] To achieve the above objectives, the present invention is implemented through the following technical solutions: a dialogue management method for an AI intelligent outbound call system, comprising the following steps:

[0012] Sampling the speech signal, performing fast Fourier transform on the speech signal of several samples to obtain the spectrum of the speech signal, using Gaussian function to approximate the frequency response of the filter, performing point multiplication operation on the fast Fourier transform result and the frequency response of the filter, and summing them to obtain the energy of each filter;

[0013] Performing logarithmic transformation on the energies of several filters, obtaining the first-order difference coefficient by calculating the difference between the logarithms of the energies of adjacent filters, obtaining the second-order difference coefficient by calculating the difference between adjacent first-order difference coefficients, and performing recursive calculation to obtain multi-order difference coefficients;

[0014] By analyzing the difference coefficient of each order, calculating the difference coefficient fluctuation index, calculating the difference coefficient fluctuation index of each order difference coefficient, and selecting the order where the difference coefficient fluctuation index has the minimum value as the order of the multi-order difference coefficient;

[0015] For each filter, the difference between the speech peak intensity and the noise intensity is calculated, and the sum is calculated to obtain the clarity index. The standard deviation of the multi-order difference coefficients is calculated through the multi-order difference coefficients. Combined with the clarity index and the signal-to-noise ratio, the speech quality evaluation value is calculated. When the speech quality evaluation value is less than the quality threshold, switch to manual outbound calling.

[0016] Furthermore, the speech signals of several samples obtained by sampling are subjected to fast Fourier transform, and the speech signals in the time domain are converted to the frequency domain to obtain the spectrum of the speech signal. The calculation formula is as follows:

[0017]

[0018] Where X[k] represents the fast Fourier transform result, k represents the frequency index, k = 1, 2..., N, N is a positive integer, x[n] is the nth sample in the time domain, n = 1, 2..., N, N is a positive integer, j is the imaginary unit, and j 2 =-1.

[0019] Furthermore, the frequency response of the Gammatone filter is approximated by a Gaussian function, and the calculation formula is as follows:

[0020]

[0021] Among them, H m [k] represents the frequency response of the filter, f[k] represents the kth frequency point in the frequency array, and f c is the center frequency of the filter and σ is the standard deviation of the Gaussian function.

[0022] Furthermore, the fast Fourier transform result is multiplied by the frequency response of the Gammatone filter and summed to obtain the energy of each filter. The calculation formula is as follows:

[0023]

[0024] Among them, E m represents the energy of the mth Gammatone filter, X[k] represents the fast Fourier transform result, H m [k represents the frequency response of the Gammatone filter, k represents the frequency index, k=1,2…,N, N is a positive integer.

[0025] Furthermore, the energies of several filters are logarithmically transformed, and the first-order difference coefficient is obtained by calculating the difference between the logarithms of the energies of adjacent filters, and the second-order difference coefficient is obtained by calculating the difference between adjacent first-order difference coefficients. The calculation formula is as follows:

[0026]

[0027] Where, ΔlogE m represents the first-order coefficient of difference, Δ 2 logE m represents the second-order coefficient of variation, E m Represents the energy of the mth Gammatone filter.

[0028] Furthermore, recursive calculation is performed to obtain multi-order difference coefficients, and the calculation formula is as follows:

[0029] Δ p logE m =Δ p-1 logE m+1 -Δ p-1 logE m

[0030] Among them, Δ p logE m represents the multi-order difference coefficient, and p represents the order.

[0031] Furthermore, by analyzing the difference coefficient of each order, the difference coefficient fluctuation index is calculated, and the calculation formula is as follows:

[0032]

[0033] Among them, MDV represents the variance coefficient volatility index, Δ p logE m represents the mth p-order difference coefficient, m=1,2,…,M, M is a positive integer, μ represents the mean of m p-order difference coefficients;

[0034] The difference coefficient fluctuation index of each order difference coefficient is calculated, and the order where the difference coefficient fluctuation index has the minimum value is selected as the order of the multi-order difference coefficient.

[0035] Furthermore, for each filter, the speech peak and noise intensity are measured, the difference between the speech peak intensity and the noise intensity is calculated, and the sum is calculated to obtain the clarity index. The calculation formula is as follows:

[0036]

[0037] Where CI represents the clarity index, P m Represents the speech peak intensity of the mth filter, NI m Represents the noise intensity of the mth filter, m = 1, 2, …, M, where M is a positive integer.

[0038] Furthermore, the multi-order difference coefficients with determined orders are obtained and the standard deviation of the multi-order difference coefficients is calculated. The voice quality evaluation value is calculated by combining the clarity index and the signal-to-noise ratio. The calculation formula is as follows:

[0039]

[0040] Where VQ represents the voice quality evaluation value, Δ p logE m represents the mth p-order difference coefficient, μ represents the mean of the p-order difference coefficient, CI represents the clarity index, SNR represents the signal-to-noise ratio, k1, k2 and k3 represent weight coefficients, 0 <k1<1,0<k2<1,0<k3<1。

[0041] A dialogue management system for an AI intelligent outbound call system, comprising: a voice signal processing module, a multi-order difference coefficient generation module, a difference coefficient selection module and a voice quality assessment module; wherein:

[0042] The speech signal processing module samples the speech signal, performs fast Fourier transform on the speech signals of several samples to obtain the spectrum of the speech signal, uses Gaussian function to approximate the frequency response of the filter, performs point multiplication operation on the fast Fourier transform result and the frequency response of the filter, and sums them to obtain the energy of each filter;

[0043] A multi-order difference coefficient generation module performs logarithmic transformation on the energy of several filters, obtains the first-order difference coefficient by calculating the difference between the logarithms of the energies of adjacent filters, obtains the second-order difference coefficient by calculating the difference between adjacent first-order difference coefficients, and performs recursive calculation to obtain the multi-order difference coefficients;

[0044] The difference coefficient selection module calculates the difference coefficient fluctuation index by analyzing the difference coefficient of each order, calculates the difference coefficient fluctuation index of each order difference coefficient, and selects the order where the difference coefficient fluctuation index has the minimum value as the order of the multi-order difference coefficient;

[0045] The dialogue management module calculates the difference between the speech peak intensity and the noise intensity for each filter, and sums them up to obtain the clarity index. The standard deviation of the multi-order difference coefficients is calculated through the multi-order difference coefficients. The clarity index and the signal-to-noise ratio are combined to calculate the speech quality assessment value. When the speech quality assessment value is less than the quality threshold, it switches to manual outbound calling.

[0046] (III) Beneficial effects

[0047] The present invention provides a dialogue management method and system for an AI intelligent outbound call system, which has the following beneficial effects:

[0048] (1) By sampling and fast Fourier transforming speech signals, we can obtain the spectrum of speech signals, which helps to better understand and analyze the characteristics of speech signals. Further, by using Gaussian functions to approximate the frequency response of the Gammatone filter, we can more accurately simulate the human ear's perception of sound, thereby extracting more critical and effective speech features for subsequent speech recognition, speech synthesis and other tasks, thereby improving the performance and accuracy of the system.

[0049] (2) By performing logarithmic transformation on the filter energy and calculating the coefficient of difference, the non-stationarity in the data can be effectively eliminated, making the data more stable. This is particularly important for the dialogue management of the AI ​​intelligent outbound call system, because stable data helps the system to more accurately identify and analyze key information in the dialogue.

[0050] (3) By conducting an in-depth analysis of the difference coefficient of each order and calculating its difference coefficient fluctuation index, we can find the order that can most stably and accurately reflect the characteristics of the voice signal. The smaller the difference coefficient fluctuation index, the smaller the range of variation of the difference coefficient of this order and the higher the stability. Therefore, it is more suitable as the order of multi-order difference coefficients, which helps to reduce the error and uncertainty of the AI ​​intelligent outbound call system, thereby improving the overall stability and reliability of the system.

[0051] (4) By combining multiple indicators such as multi-order difference coefficient, clarity index and signal-to-noise ratio to accurately evaluate voice quality, it is possible to more accurately determine when it is necessary to switch to manual outbound calling. By combining multiple voice quality evaluation indicators and automatic switching mechanisms, the intelligence level of the AI ​​intelligent outbound calling system has been improved, which can better cope with complex and changing voice environments and provide users with more intelligent and efficient services. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a schematic diagram of the steps of the dialogue management method of the AI ​​intelligent outbound calling system of the present invention;

[0053] Figure 2 This is a flow chart of a conversation management method of an AI intelligent outbound calling system of the present invention;

[0054] Figure 3 The figure is a structural diagram of the dialogue management system of the AI ​​intelligent outbound calling system of the present invention. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] See also Figure 1-Figure 2 The present invention provides a dialogue management method for an AI intelligent outbound calling system, comprising the following steps:

[0057] Step 1: Sample the speech signal, perform fast Fourier transform on the speech signal of several samples to obtain the spectrum of the speech signal, use Gaussian function to approximate the frequency response of the filter, perform point multiplication operation on the fast Fourier transform result and the frequency response of the filter, and sum them to obtain the energy of each filter;

[0058] The step 1 includes the following contents:

[0059] Step 101: Perform voice interaction with the user through the AI ​​intelligent outbound call system to obtain voice data during the conversation;

[0060] Step 102: setting a sampling frequency, sampling the acquired voice signal according to the set sampling frequency, discretizing the continuous voice signal, and converting it into a series of discrete sample values;

[0061] It should be noted that in the intelligent outbound call system, there is no fixed standard value for the sampling rate setting, because it depends on many factors, including the required sound quality, storage space, processing speed, and overall system performance. Generally speaking, a higher sampling rate can provide better sound quality, but it will also take up more storage space and computing resources. Therefore, when selecting the sampling rate, it is necessary to weigh the actual needs and system performance. For general intelligent outbound call systems, commonly used sampling rates include 16kHz, 32kHz, and 44.1kHz, etc. These sampling rates can ensure call quality while maintaining relatively reasonable storage and processing requirements;

[0062] Step 103: Perform fast Fourier transform on the speech signals of the sampled samples, convert the speech signals in the time domain into the frequency domain, and obtain the spectrum of the speech signals. The calculation formula is as follows:

[0063]

[0064] Where X[k] represents the fast Fourier transform result, k represents the frequency index, k = 1, 2…, N, N is a positive integer, x[n] is the nth sample in the time domain, n = 1, 2…, N, N is a positive integer, j is the imaginary unit, and j 2 = -1;

[0065] Step 104: Use the Gaussian function to approximate the frequency response of the Gammatone filter. The calculation formula is as follows:

[0066]

[0067] Among them, H m [k] represents the frequency response of the filter, f[k] represents the kth frequency point in the frequency array, and f c is the center frequency of the filter, σ is the standard deviation of the Gaussian function, and the relationship between σ and bandwidth is:

[0068]

[0069] Wherein, bw represents bandwidth;

[0070] It should be noted that the Gammatone filter is a filter widely used in auditory modeling and speech signal processing. It simulates the response characteristics of the basilar membrane of the human ear to sounds of different frequencies. The main feature of the Gammatone filter is its impulse response shape, which combines the characteristics of the Gamma distribution and the sine wave, thus being able to simulate the human ear's perception of sound well.

[0071] Step 105: Perform a dot multiplication operation on the fast Fourier transform result and the frequency response of the Gammatone filter, and sum them to obtain the energy of each filter. The calculation formula is as follows:

[0072]

[0073] Among them, E m represents the energy of the mth Gammatone filter, X[k] represents the fast Fourier transform result, H m [k represents the frequency response of the Gammatone filter, k represents the frequency index, k=1,2…,N, N is a positive integer.

[0074] It should be noted that performing a dot product operation on the FFT result (i.e., spectrum) and the frequency response of the Gammatone filter means multiplying the amplitude value on each frequency component with the filter response value at the corresponding frequency, so that the amplitude value of each frequency component after filter weighting can be obtained.

[0075] When using, combine the contents of step 101 to step 105:

[0076] By sampling and fast Fourier transforming the speech signal, the spectrum of the speech signal can be obtained, which helps to better understand and analyze the characteristics of the speech signal. Further using the Gaussian function to approximate the frequency response of the Gammatone filter can more accurately simulate the human ear's perception of sound, thereby extracting more critical and effective speech features for subsequent speech recognition, speech synthesis and other tasks, improving the performance and accuracy of the system.

[0077] Step 2: Logarithmically transform the energies of several filters, obtain the first-order difference coefficient by calculating the difference between the logarithms of the energies of adjacent filters, obtain the second-order difference coefficient by calculating the difference between adjacent first-order difference coefficients, and perform recursive calculation to obtain multi-order difference coefficients;

[0078] The step 2 includes the following contents:

[0079] Step 201: Logarithmically transform the energies of several filters, obtain the first-order difference coefficient by calculating the difference between the logarithms of adjacent filter energies, and obtain the second-order difference coefficient by calculating the difference between adjacent first-order difference coefficients. The calculation formula is as follows:

[0080]

[0081] Where, ΔlogE m represents the first-order coefficient of difference, Δ 2 logE m represents the second-order coefficient of variation, E m Represents the energy of the mth Gammatone filter;

[0082] Step 202: Perform recursive calculation to obtain multi-order difference coefficients, and the calculation formula is as follows:

[0083] Δ p logE m =Δ p-1 logE m+1 -Δ p-1 logE m

[0084] Among them, Δ p logE m represents the multi-order difference coefficient, and p represents the order.

[0085] It should be noted that after obtaining the energy of each filter, in order to further simplify the feature representation and improve robustness, it is necessary to take the logarithm of these energy values. The energy values ​​of audio signals may span a large range. Directly using these original values ​​may cause some algorithms or models to be overly sensitive to larger energy values ​​and insensitive to smaller energy values. By taking the logarithm of the energy value, the dynamic range can be compressed, making energy values ​​of different sizes closer in numerical value, thereby improving the stability of the algorithm; the human ear's perception of sound is not linear, but logarithmic. Therefore, taking the logarithm of the energy value can make the feature representation more consistent with the auditory characteristics of the human ear, thereby improving the performance of the speech or audio processing system; in some cases, the energy value after taking the logarithm may be closer to the normal distribution, which helps to simplify subsequent statistical analysis and model training.

[0086] When used, combine step 201 to step 202:

[0087] By performing logarithmic transformation on the filter energy and calculating the coefficient of difference, the non-stationarity in the data can be effectively eliminated, making the data more stable. This is especially important for the dialogue management of AI intelligent outbound call systems, because stable data helps the system to more accurately identify and analyze key information in the dialogue.

[0088] Step 3: Analyze the difference coefficient of each order, calculate the difference coefficient fluctuation index, calculate the difference coefficient fluctuation index of each order difference coefficient, and select the order where the difference coefficient fluctuation index has the minimum value as the order of the multi-order difference coefficient;

[0089] The step three includes the following contents:

[0090] Step 301: Analyze the difference coefficient of each order and calculate the difference coefficient fluctuation index. The calculation formula is as follows:

[0091]

[0092] Among them, MDV represents the variance coefficient volatility index, Δ p logE m represents the mth p-order difference coefficient, m=1,2,…,M, M is a positive integer, μ represents the mean of m p-order difference coefficients;

[0093] Step 302: Calculate the difference coefficient fluctuation index of each order difference coefficient, and select the order where the difference coefficient fluctuation index has the minimum value as the order of the multi-order difference coefficient.

[0094] It should be noted that the order where the difference coefficient fluctuation index has the minimum value is selected as the order of the multi-order difference coefficient based on the consideration of stability and reliability. Too low an order may not be able to fully capture the complex characteristics of the speech signal, while too high an order may introduce too much noise and unstable factors. Therefore, by comparing the difference coefficient fluctuation index of different orders, a balance point can be found that can fully reflect the characteristics of the speech signal while maintaining high stability and reliability.

[0095] When used, combine the contents of step 301 to step 302:

[0096] By conducting an in-depth analysis of the difference coefficient of each order and calculating its difference coefficient fluctuation index, we can find the order that can most stably and accurately reflect the characteristics of the voice signal. The smaller the difference coefficient fluctuation index, the smaller the range of variation of the difference coefficient of this order and the higher the stability. Therefore, it is more suitable as the order of multi-order difference coefficients, which helps to reduce the error and uncertainty of the AI ​​intelligent outbound call system, thereby improving the overall stability and reliability of the system.

[0097] Step 4: For each filter, calculate the difference between the speech peak intensity and the noise intensity, and sum them up to obtain the clarity index. Calculate the standard deviation of the multi-order difference coefficients through the multi-order difference coefficients. Combine the clarity index and the signal-to-noise ratio to calculate the speech quality assessment value. When the speech quality assessment value is less than the quality threshold, switch to manual outbound calling.

[0098] The step 4 includes the following contents:

[0099] Step 401: for each filter, measure the speech peak (i.e., the strength of the speech signal) and the strength of the noise, calculate the difference between the speech peak strength and the noise strength, and sum them up to obtain the clarity index. The calculation formula is as follows:

[0100]

[0101] Where CI represents the clarity index, P m Represents the speech peak intensity of the mth filter, NI m represents the noise intensity of the mth filter, m = 1, 2, ..., M, M is a positive integer;

[0102] Step 402: Obtain the multi-order difference coefficients with determined orders and calculate the standard deviation of the multi-order difference coefficients, and calculate the speech quality evaluation value in combination with the clarity index and the signal-to-noise ratio. The calculation formula is as follows:

[0103]

[0104] Where VQ represents the voice quality evaluation value, Δ p logE m represents the mth p-order difference coefficient, μ represents the mean of the p-order difference coefficient, CI represents the clarity index, SNR represents the signal-to-noise ratio, k1, k2 and k3 represent weight coefficients, 0 <k1<1,0<k2<1,0<k3<1;

[0105] Step 403: pre-set a quality threshold. When the voice quality evaluation value is less than the quality threshold, it indicates that the current voice quality is poor and the manual outbound call is switched.

[0106] It should be noted that the signal-to-noise ratio (SNR) is usually defined as the ratio of signal power to noise power, and the calculation formula is as follows: Among them, P signal Represents the signal power, P noise Indicates the noise power. The higher the SNR, the better the voice quality, and vice versa.

[0107] When used, combine step 401 to step 403:

[0108] By combining multiple indicators such as multi-order difference coefficient, clarity index and signal-to-noise ratio to accurately evaluate voice quality, it is possible to more accurately determine when it is necessary to switch to manual outbound calling. By combining multiple voice quality evaluation indicators and automatic switching mechanisms, the intelligence level of the AI ​​intelligent outbound calling system has been improved, which can better cope with complex and changing voice environments and provide users with more intelligent and efficient services.

[0109] See also Figure 3 The present invention also provides a dialogue management system for an AI intelligent outbound call system, comprising: a voice signal processing module, a multi-order difference coefficient generation module, a difference coefficient selection module and a voice quality evaluation module; wherein,

[0110] The speech signal processing module samples the speech signal, performs fast Fourier transform on the speech signals of several samples to obtain the spectrum of the speech signal, uses Gaussian function to approximate the frequency response of the filter, performs point multiplication operation on the fast Fourier transform result and the frequency response of the filter, and sums them to obtain the energy of each filter;

[0111] A multi-order difference coefficient generation module performs logarithmic transformation on the energy of several filters, obtains the first-order difference coefficient by calculating the difference between the logarithms of the energies of adjacent filters, obtains the second-order difference coefficient by calculating the difference between adjacent first-order difference coefficients, and performs recursive calculation to obtain the multi-order difference coefficients;

[0112] The difference coefficient selection module calculates the difference coefficient fluctuation index by analyzing the difference coefficient of each order, calculates the difference coefficient fluctuation index of each order difference coefficient, and selects the order where the difference coefficient fluctuation index has the minimum value as the order of the multi-order difference coefficient;

[0113] The dialogue management module calculates the difference between the speech peak intensity and the noise intensity for each filter, and sums them up to obtain the clarity index. The standard deviation of the multi-order difference coefficients is calculated through the multi-order difference coefficients. The clarity index and the signal-to-noise ratio are combined to calculate the speech quality assessment value. When the speech quality assessment value is less than the quality threshold, it switches to manual outbound calling.

[0114] In the application, the several formulas involved are all calculated by removing dimensions and taking their numerical values. The formula is a formula of the most recent real situation obtained by collecting a large amount of data and performing software simulation. The coefficients in the formula are set by technicians in this field according to actual conditions.

[0115] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in combination with electronic hardware, computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0116] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0117] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A dialogue management method for an AI intelligent outbound call system, characterized in that: The following steps are involved: Sampling the speech signal, performing fast Fourier transform on the speech signal of several samples to obtain the spectrum of the speech signal, using Gaussian function to approximate the frequency response of the filter, performing point multiplication operation on the fast Fourier transform result and the frequency response of the filter, and summing them to obtain the energy of each filter; Performing logarithmic transformation on the energies of several filters, obtaining the first-order difference coefficient by calculating the difference between the logarithms of the energies of adjacent filters, obtaining the second-order difference coefficient by calculating the difference between adjacent first-order difference coefficients, and performing recursive calculation to obtain multi-order difference coefficients; By analyzing the difference coefficient of each order, calculating the difference coefficient fluctuation index, calculating the difference coefficient fluctuation index of each order difference coefficient, and selecting the order where the difference coefficient fluctuation index has the minimum value as the order of the multi-order difference coefficient; For each filter, the difference between the speech peak intensity and the noise intensity is calculated, and the sum is calculated to obtain the clarity index. The standard deviation of the multi-order difference coefficients is calculated through the multi-order difference coefficients. Combined with the clarity index and the signal-to-noise ratio, the speech quality evaluation value is calculated. When the speech quality evaluation value is less than the quality threshold, switch to manual outbound calling.

2. According to claim 1, a dialogue management method for an AI intelligent outbound call system is characterized in that: Perform fast Fourier transform on the speech signals of several samples obtained by sampling, convert the speech signal in the time domain into the frequency domain, and obtain the spectrum of the speech signal. The calculation formula is as follows: Where X[k] represents the fast Fourier transform result, k represents the frequency index, k = 1, 2..., N, N is a positive integer, x[n] is the nth sample in the time domain, n = 1, 2..., N, N is a positive integer, j is the imaginary unit, and j 2 =-1.

3. According to claim 2, a dialogue management method for an AI intelligent outbound call system is characterized in that: Use the Gaussian function to approximate the frequency response of the Gammatone filter. The calculation formula is as follows: Among them, H m [k] represents the frequency response of the filter, f[k] represents the kth frequency point in the frequency array, and f c is the center frequency of the filter and σ is the standard deviation of the Gaussian function.

4. According to claim 3, a dialogue management method for an AI intelligent outbound call system is characterized in that: Perform a dot multiplication operation on the fast Fourier transform result and the frequency response of the gammatone filter and sum them to get the energy of each filter. The calculation formula is as follows: Among them, E m represents the energy of the mth Gammatone filter, X[k] represents the fast Fourier transform result, H m [k represents the frequency response of the Gammatone filter, k represents the frequency index, k=1,2...,N, N is a positive integer.

5. The dialogue management method of an AI intelligent outbound call system according to claim 1, characterized in that: The energy of several filters is logarithmically transformed, and the first-order difference coefficient is obtained by calculating the difference between the logarithms of the energy of adjacent filters, and the second-order difference coefficient is obtained by calculating the difference between adjacent first-order difference coefficients. The calculation formula is as follows: Where, ΔlogE m represents the first-order coefficient of difference, Δ 2 logE m represents the second-order coefficient of variation, E m Represents the energy of the mth Gammatone filter.

6. The method for dialogue management of an AI intelligent outbound call system according to claim 5, characterized in that: Recursive calculation is performed to obtain multi-order difference coefficients. The calculation formula is as follows: Δ p logE m =Δ p-1 logE m+1 -Δ p-1 logE m Among them, Δ p logE m represents the multi-order difference coefficient, and p represents the order.

7. The dialogue management method of an AI intelligent outbound call system according to claim 1, characterized in that: By analyzing the difference coefficient of each order, the difference coefficient fluctuation index is calculated. The calculation formula is as follows: Among them, MDV represents the variance coefficient volatility index, Δ p logE m represents the mth p-order difference coefficient, m=1,2,...,M, M is a positive integer, μ represents the mean of the p-order difference coefficient; The difference coefficient fluctuation index of each order difference coefficient is calculated, and the order where the difference coefficient fluctuation index has the minimum value is selected as the order of the multi-order difference coefficient.

8. The method for dialogue management of an AI intelligent outbound call system according to claim 1, characterized in that: For each filter, the speech peak and noise intensity are measured, the difference between the speech peak intensity and the noise intensity is calculated, and the sum is calculated to obtain the clarity index. The calculation formula is as follows: Where CI represents the clarity index, P m Represents the speech peak intensity of the mth filter, NI m Represents the noise intensity of the mth filter, m=1,2,...,M, where M is a positive integer.

9. The method for dialogue management of an AI intelligent outbound calling system according to claim 8, characterized in that: The multi-order difference coefficients with determined orders are obtained and the standard deviation of the multi-order difference coefficients is calculated. The voice quality evaluation value is calculated by combining the clarity index and the signal-to-noise ratio. The calculation formula is as follows: Where VQ represents the voice quality evaluation value, Δ p logE m represents the mth p-order difference coefficient, μ represents the mean of the p-order difference coefficient, CI represents the clarity index, SNR represents the signal-to-noise ratio, k1, k2 and k3 represent weight coefficients, 0 <k1<1,0<k2<1,0<k3<1。 10. A dialogue management system for an AI intelligent outbound call system, used to implement the method described in any one of claims 1 to 9, characterized in that: include: The speech signal processing module samples the speech signal, performs fast Fourier transform on the speech signals of several samples to obtain the spectrum of the speech signal, uses Gaussian function to approximate the frequency response of the filter, performs point multiplication operation on the fast Fourier transform result and the frequency response of the filter, and sums them to obtain the energy of each filter; A multi-order difference coefficient generation module performs logarithmic transformation on the energy of several filters, obtains the first-order difference coefficient by calculating the difference between the logarithms of the energies of adjacent filters, obtains the second-order difference coefficient by calculating the difference between adjacent first-order difference coefficients, and performs recursive calculation to obtain the multi-order difference coefficients; The difference coefficient selection module calculates the difference coefficient fluctuation index by analyzing the difference coefficient of each order, calculates the difference coefficient fluctuation index of each order difference coefficient, and selects the order where the difference coefficient fluctuation index has the minimum value as the order of the multi-order difference coefficient; The dialogue management module calculates the difference between the speech peak intensity and the noise intensity for each filter, and sums them up to obtain the clarity index. The standard deviation of the multi-order difference coefficients is calculated through the multi-order difference coefficients. The clarity index and the signal-to-noise ratio are combined to calculate the speech quality assessment value. When the speech quality assessment value is less than the quality threshold, it switches to manual outbound calling.

Images