Voiceprint waveform visual extraction and matching degree calculation method

Abstract

The invention discloses a voiceprint waveform visual extraction and matching degree calculation method, and relates to the technical field of biological feature recognition, and the method comprises the following specific steps: voiceprint data collection: collecting a voiceprint signal of a target object through a recording device, and obtaining original voiceprint data; voiceprint preprocessing: carrying out denoising and normalization processing on the original voiceprint data, and removing environmental interference and redundant information to obtain a standardized voiceprint signal; according to the method, the matching model is constructed based on the inherent correlation characteristics of the trigonometric function, the amplitude quantification information of the voiceprint characteristics is fully utilized, the characteristic matching degree of the two groups of voiceprints is comprehensively described through the comprehensive operation of the multi-frequency point matching coefficients, the matching degree calculation accuracy is improved, the misjudgment rate in a complex scene is reduced, and meanwhile, the accuracy of matching degree calculation is improved. The matching degree threshold value can be dynamically adjusted according to the precision requirements of different application scenes such as target crowd searching, biological feature anti-certification and identity verification, and various requirements can be flexibly met.

Description

Technical Field

[0001] This invention relates to the field of biometric recognition technology, specifically a method for visual extraction and matching degree calculation of voiceprint waveforms. Background Technology

[0002] Voiceprint, as one of the unique biometric features of the human body, has been widely used in many fields such as identity recognition and security verification due to its significant advantages such as strong stability and convenient collection. With the continuous development and progress of science and technology, the application demand of voiceprint recognition technology in high-precision scenarios such as target group search, biometric verification, and identity verification is increasing. This places higher demands on the visualization effect and matching accuracy of voiceprint recognition technology. Voiceprint signals are essentially signals that change continuously over time in the time domain, containing rich feature information. How to extract representative and distinguishable features from these complex signals and present them in an intuitive way, while achieving high-precision matching calculations, has become a key problem that urgently needs to be solved in the field of voiceprint recognition technology.

[0003] Most existing voiceprint recognition technologies rely on direct comparison of voiceprint feature parameters. This approach has significant limitations. First, it performs poorly in terms of visualization. Traditional methods often only focus on the numerical features of voiceprints, lacking an intuitive presentation method. This makes the expression of voiceprint features not intuitive and vivid enough. In complex voiceprint comparison processes, it is difficult for staff to quickly and accurately grasp the differences and commonalities between voiceprints through simple numerical comparisons. Once an anomaly in the matching result occurs, it is also difficult to trace the matching process and find the root cause of the problem. Second, it is insufficient in the accuracy of matching degree calculation. Traditional matching methods often use Euclidean distance, correlation analysis, and other means. Although these methods can measure the similarity between voiceprint features to a certain extent, they are not accurate enough in characterizing the fit of voiceprint features. In practical application scenarios, environmental noise, equipment differences, and changes in individual vocal states can all affect voiceprint features. Traditional methods are difficult to effectively deal with these complex situations, leading to misjudgments and omissions in complex scenarios. This greatly limits the use of voiceprint technology in finding target groups and verifying biometric features. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method for voiceprint waveform visualization extraction and matching degree calculation. This method ensures the integrity and accuracy of the original data by using recording equipment to collect the voiceprint signal of the target object in a quiet, interference-free environment during the voiceprint data acquisition stage. In the voiceprint preprocessing stage, signal denoising technology is used to remove environmental noise, and normalization processing is used to eliminate amplitude differences caused by different acquisition devices and distances. The signal is also segmented and amplitude calibrated to ensure the stability and consistency of the standardized voiceprint signal. During the voiceprint visualization extraction process, a mathematical transformation method based on the Fourier transform principle is employed. This method converts time-domain voiceprint signals into frequency-domain features, extracts key feature parameters such as amplitude, frequency, and phase, and presents the voiceprint waveform intuitively in the form of a spectrum or waveform curve. In terms of matching degree calculation, a matching model is constructed based on the inherent correlation characteristics of trigonometric functions. The feature parameters of two sets of visualized voiceprints are quantitatively compared and calculated. Through the comprehensive calculation of the matching coefficients at multiple frequency points, the feature matching degree of the two sets of voiceprints is fully characterized, and an accurate matching degree value is output. Finally, the matching degree threshold is dynamically set according to the accuracy requirements of different application scenarios to make accurate matching judgments, providing new ideas and solutions for the development and application of voiceprint recognition technology.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for visual extraction and matching degree calculation of voiceprint waveforms, the method comprising the following specific steps:

[0006] Voiceprint data acquisition: Acquire the voiceprint signal of the target object through recording equipment to obtain raw voiceprint data;

[0007] Voiceprint preprocessing: The original voiceprint data is denoised and normalized to remove environmental interference and redundant information, and a standardized voiceprint signal is obtained.

[0008] Voiceprint visualization extraction: A mathematical transformation method based on the Fourier transform principle is used to convert the standardized voiceprint signal, map the time-domain voiceprint signal to frequency-domain features, extract amplitude, frequency and phase parameters, and generate a visualized voiceprint waveform.

[0009] Matching degree calculation: Extract feature parameters from two sets of visualized voiceprint waveforms, construct a matching model based on the inherent correlation characteristics of trigonometric functions, and calculate the feature fit of the two sets of voiceprints through model quantification to obtain the matching degree value;

[0010] Matching determination: Set a matching threshold, and determine whether the two sets of voiceprints match based on the comparison between the calculated matching value and the threshold.

[0011] Furthermore, in the voiceprint preprocessing step, signal denoising technology is used to remove environmental noise from the original voiceprint data, retain the core features of the voiceprint, normalize the amplitude of the denoised voiceprint signal to the [0,1] interval, eliminate amplitude differences caused by different acquisition devices and acquisition distances, ensure the comparability of voiceprint signals from different sources, segment the normalized voiceprint signal, and perform amplitude calibration to reduce the impact of local signal fluctuations.

[0012] Furthermore, in the voiceprint visualization extraction step, a mathematical transformation method based on the Fourier transform principle is used to convert the standardized voiceprint signal. The specific steps are as follows: When performing the conversion operation on the preprocessed standardized voiceprint signal using the mathematical transformation method based on the Fourier transform principle, the continuous time-domain voiceprint signal is first discretized. Combining the frequency characteristics of the voiceprint signal and the sampling frequency during acquisition, the time window length is set, and the standardized voiceprint signal is divided into several signal segments of uniform length and stable characteristics. For each signal segment, relying on the Fourier transform, the amplitude change information that fluctuates continuously with time in the time dimension is mapped to the frequency dimension, and the time-domain voiceprint signal is decomposed into a series of sine wave superpositions with different frequencies and amplitudes. Each set of sine waves corresponds to a frequency component of the voiceprint signal. During the conversion process, the amplitude and phase information corresponding to each frequency component are calculated to clarify the energy distribution characteristics of the voiceprint signal in different frequency ranges. After the conversion is completed, the obtained frequency domain signal is presented in the form of quantized frequency-amplitude and frequency-phase data pairs.

[0013] Furthermore, in the voiceprint visualization extraction step, the time-domain voiceprint signal is mapped to frequency-domain features, and amplitude, frequency, and phase parameters are extracted. The specific steps are as follows: When extracting amplitude parameters, for the signal energy value corresponding to each frequency point in the frequency domain space, peak detection and effective value calculation are used to record the amplitude peak, mean, and variation gradient of different frequency intervals; when extracting frequency parameters, the effective frequency range of the voiceprint signal is focused, the fundamental frequency position and harmonic frequency interval are located through energy distribution analysis, key frequency points corresponding to resonance peaks are marked, and the distribution law of each frequency component is clarified; when extracting phase parameters, the initial phase and phase offset of each frequency component are captured, the phase change law of different frequency components during propagation is recorded, and the feature dimensions that amplitude parameters cannot cover are supplemented; after the parameters are extracted, a visualized voiceprint waveform is generated with the frequency-amplitude and frequency-phase quantized data pairs as the core.

[0014] Furthermore, in the matching degree calculation step, feature parameters of the two sets of visualized voiceprint waveforms are extracted. The specific steps are as follows: For the two sets of visualized voiceprint waveforms to be matched, the parameter information of the corresponding feature dimensions is matched one by one according to the same frequency scale and time segmentation standard. The amplitude quantization value and phase offset corresponding to each key frequency point are extracted, and the change trend data of each parameter in the time segment are recorded. A matching model is constructed based on the inherent correlation characteristics of trigonometric functions. The quantization value of the corresponding position in the two sets of parameter sets is used as input. The amplitude complementarity and phase coordination characteristics of sine and cosine functions are used to transform the numerical difference of the two sets of parameters into a matching degree index in the trigonometric function correlation domain. Specifically, the quantization value of each set of feature parameters is mapped to a reasonable numerical range suitable for trigonometric function operation. Then, through the inherent correspondence of trigonometric functions, the correlation strength of the two sets of parameters at the same feature point is calculated, and the feature matching coefficient of a single feature point is generated. Subsequently, the matching coefficients of all corresponding feature points are weighted and comprehensively calculated to obtain a matching degree value that can comprehensively characterize the overall matching level of the two sets of voiceprints.

[0015] Furthermore, in the matching degree calculation step, the correlation strength between the two sets of parameters at the same feature point is calculated using the inherent correspondence of trigonometric functions. The calculation formula is as follows: ,in, It is the first The association strength of each feature point ranges from [0,1]. The closer the value is to 1, the higher the compatibility between the two sets of voiceprints at that feature point. The first set of voiceprints was in the... The amplitude quantization values ​​of each feature point have been normalized to [0, ] interval, The second set of voiceprints is in the... The amplitude quantization value of each feature point, and For the same frequency point, The first set of voiceprints was in the... Phase offset of each feature point The second set of voiceprints is in the... The phase offset of each feature point, and Corresponding to the same frequency point.

[0016] Furthermore, in the matching degree calculation step, a weighted comprehensive calculation is performed on the matching coefficients of all corresponding feature points, and the calculation formula is as follows: ,in, This represents the final matching degree between the two sets of voiceprints, with a value ranging from [0,1]. The closer the value is to 1, the higher the overall matching degree between the two sets of voiceprints. It is the first The weight coefficients for each feature point are set according to the importance of the feature point. It is the first The correlation strength of each feature point It is the total number of feature points involved in the matching.

[0017] Furthermore, in the matching determination step, a matching degree threshold is dynamically set according to the accuracy requirements of different application scenarios. For identity verification scenarios, set... ≥0.9; For scenarios involving finding matching users, set ≥0.8; For scenarios that disprove biometric evidence, set ≥0.85.

[0018] Furthermore, in the matching determination step, if the calculated matching degree value... ≥Preset threshold If so, then the two sets of voiceprints are determined to belong to the same object; if < If so, it is determined that the two sets of voiceprints do not belong to the same object.

[0019] Compared with existing technologies, this method for visual extraction and matching degree calculation of voiceprint waveforms has the following advantages:

[0020] I. This invention constructs a matching model based on the inherent correlation characteristics of trigonometric functions, fully utilizes the amplitude quantification information of voiceprint features, and comprehensively describes the feature matching degree of two sets of voiceprints through the comprehensive calculation of multi-frequency point matching coefficients. This improves the accuracy of matching degree calculation and reduces the false judgment rate in complex scenarios. At the same time, the matching degree threshold can be dynamically adjusted according to the accuracy requirements of different application scenarios such as target group search, biometric verification, and identity verification. It can flexibly adapt to various needs, has a wide range of applications, and is suitable for large-scale voiceprint comparison scenarios. It has strong practicality and operability.

[0021] Second, this invention uses a mathematical transformation method based on the Fourier transform principle to convert the abstract and difficult-to-understand voiceprint signal in the time domain into a visually intuitive spectrum or waveform in the frequency domain. It displays the core features of the voiceprint signal, such as frequency distribution and amplitude, in graphical form, allowing staff to clearly observe the differences and commonalities in voiceprint features. In scenarios such as identity verification and finding matching groups, staff can quickly judge voiceprint features based on the visualized graphics, and the matching process is traceable. This effectively solves the problems of poor visualization and difficulty in tracing the matching process in traditional voiceprint recognition technology, providing a more intuitive and reliable basis for the application of voiceprint recognition technology.

[0022] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0024] Figure 1 A flowchart of a method for visual extraction and matching degree calculation of voiceprint waveforms;

[0025] Figure 2 This is a flowchart of the voiceprint visualization extraction steps in a method for voiceprint waveform visualization extraction and matching degree calculation.

[0026] Figure 3 This is a flowchart of the matching degree calculation steps for a method of visual extraction and matching degree calculation of voiceprint waveforms. Detailed Implementation

[0027] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0028] This invention provides a method for voiceprint waveform visualization extraction and matching degree calculation. During the voiceprint data acquisition stage, a recording device is used to collect the voiceprint signal of the target object in a quiet environment without strong interference, ensuring the integrity and accuracy of the original data. In the voiceprint preprocessing stage, signal denoising technology is used to remove environmental noise, and normalization processing is used to eliminate amplitude differences caused by different acquisition devices and distances. The signal is also segmented and amplitude calibrated to ensure the stability and consistency of the standardized voiceprint signal. During the voiceprint visualization extraction process, a mathematical transformation method based on the Fourier transform principle is used to convert the time-domain voiceprint signal. For frequency domain features, key feature parameters such as amplitude, frequency, and phase are extracted and presented intuitively as a spectrum or waveform curve. In terms of matching degree calculation, a matching model is constructed based on the inherent correlation characteristics of trigonometric functions. The feature parameters of the two sets of visualized voiceprints are quantitatively compared and calculated. Through the comprehensive calculation of the matching coefficients at multiple frequency points, the feature matching degree of the two sets of voiceprints is fully characterized, and an accurate matching degree value is output. Finally, the matching degree threshold is dynamically set according to the accuracy requirements of different application scenarios to make accurate matching judgments, providing new ideas and solutions for the development and application of voiceprint recognition technology.

[0029] Example 1

[0030] This embodiment uses identity verification in the remote account opening business of financial institutions as an application scenario, such as... Figure 1 As shown, during the voiceprint data acquisition phase, a quiet, noise-free environment at the user's home was selected for collection. The user's voiceprint signal was collected using the built-in recording function of the financial institution's APP. The sampling frequency was set to 24kHz to ensure the integrity of the voiceprint features. During the acquisition, the user was prompted to clearly read the specified verification phrase. The recording was repeated 3 times, with a total duration of 30 seconds. Invalid segments with excessively long pauses or low volumes were automatically filtered out, and 25 seconds of valid voiceprint data were retained as the original voiceprint data. At the same time, auxiliary information such as the model of the acquisition device, the acquisition time, and the ambient noise level were recorded to ensure data traceability.

[0031] In the voiceprint preprocessing stage, the collected raw voiceprint data undergoes full-process optimization. First, signal denoising technology is used to remove interference noise such as slight current noise, wind noise outside the window, and distant background voices, while retaining the core voiceprint information that reflects the user's vocal characteristics. Then, the amplitude of the denoised voiceprint signal is normalized to the [0,1] interval to eliminate amplitude fluctuations caused by differences in the sensitivity of different mobile phone microphones and the distance between the user and the mobile phone, ensuring the comparability of the signals. Finally, the normalized voiceprint signal is divided into 25 segments according to the standard of 1s / segment, and the amplitude of each segment is calibrated to ensure that the amplitude fluctuation of each segment does not exceed 5%, ultimately obtaining a standardized voiceprint signal with strong stability and high consistency.

[0032] In the voiceprint visualization extraction stage, such as Figure 2 As shown, the standardized voiceprint signal is first converted using a mathematical transformation based on the Fourier transform principle. The continuous time-domain voiceprint signal is first discretized, and an appropriate time window length is set based on a 24kHz sampling frequency, dividing the standardized voiceprint signal into several characteristically stable segments. Then, through mathematical transformation, the amplitude variation information in the time dimension is mapped to the frequency dimension, decomposing the complex time-domain signal into a superposition of sine waves with different frequencies and amplitudes. This accurately captures the fundamental frequency, three main resonants, and corresponding harmonic frequency and amplitude data that characterize the uniqueness of the user's voiceprint, while simultaneously recording the phase offset of each frequency component. Based on the conversion... The obtained frequency domain quantization data is used to extract the peak amplitude, mean amplitude, gradient, specific frequency positions of the fundamental frequency and formants, and core feature parameters such as the initial phase and phase change law of each frequency component. Then, with 500-8000Hz as the effective frequency range, a spectrum diagram with frequency on the horizontal axis and amplitude on the vertical axis is generated. The energy intensity of each frequency point is presented intuitively through different color gradients. At the same time, the positions of key features such as the fundamental frequency and formants are marked, and a time-frequency waveform curve is generated to show the frequency change trend of the voiceprint in the time dimension. Finally, a visualized benchmark voiceprint waveform diagram containing 200 key feature points is formed.

[0033] During the matching degree calculation phase, when a user initiates a remote account opening application, the user's real-time voiceprint signal is collected according to the same requirements as the baseline voiceprint collection. This signal undergoes the same preprocessing and visualization extraction process to obtain a set of voiceprint feature parameters to be verified, containing 200 corresponding key feature points. This ensures that the feature parameters of the two sets of voiceprints correspond completely one-to-one in frequency range, time dimension, and feature type. Figure 3 As shown, the matching model built based on the inherent correlation characteristics of trigonometric functions first uses the formula for calculating the correlation strength of feature points. For each corresponding feature point, the calculation is performed, where As a reference voiceprint in the first The amplitude quantization value of each feature point For voiceprints to be verified in the first The amplitude quantization value of each feature point As a reference voiceprint in the first Phase offset of each feature point For voiceprints to be verified in the first The phase shift of each feature point is calculated using this formula to comprehensively consider the fit between the two sets of voiceprints in both amplitude and phase dimensions, yielding a fit coefficient for each feature point. The formula particularly emphasizes the weights of the fundamental frequency and the 30 key feature points corresponding to the three main formants. The weight is set to 1.5, and the weights of the remaining 170 non-critical feature points are... Set it to 1.0, and then calculate the weighted comprehensive matching degree using the formula. The matching coefficients of all feature points are calculated comprehensively (in the formula). The total number of feature points participating in the matching is 200. For the first The system integrates the matching coefficients of individual feature points, comprehensively considering the matching of amplitude and phase features, to obtain the overall matching degree value of the two sets of voiceprints. During the matching determination stage, considering the stringent security and accuracy requirements of identity verification in financial transactions, the matching degree threshold is set to 0.9. The matching degree value between the real-time voiceprint and the baseline voiceprint calculated above is 0.93, which is higher than the preset threshold. Therefore, it is determined that the two sets of voiceprints belong to the same user, and the identity verification is passed. The financial institution system allows the user to continue the remote account opening process. If the matching degree value is lower than 0.9, the identity verification is deemed unsuccessful, and the system prompts the user to re-collect the voiceprint or select another identity verification method. The entire implementation process requires no complex equipment support, is convenient and efficient, and the matching results are accurate and reliable, effectively preventing identity fraud risks and ensuring the safe operation of financial transactions.

[0034] Example 2

[0035] This embodiment uses voiceprint access control in the core area of ​​an enterprise as an application scenario. During the voiceprint data collection phase, authorized personnel voiceprints are collected in a recording studio designated by the enterprise's security department, with the ambient noise level controlled below 28dB. Professional recording equipment uniformly configured by the enterprise is used. Authorized personnel are prompted to clearly read the security verification phrase, and the recording is repeated 3 times for a total duration of 28 seconds. Technicians select 24 seconds of valid voiceprint data with no pronunciation deviation and stable volume as the original authorized voiceprint data, and simultaneously record the device number, collection time, and authorized personnel identity information and archive them. Unauthorized personnel voiceprints are collected at a temporary collection point next to the core area access control, in a quiet corner of the office area. The same model of recording equipment as the authorized voiceprint collection is used, and the sampling frequency is also set to 22kHz. The personnel are prompted to read the same security verification phrase, and the recording is repeated 2 times for a total duration of 22 seconds. The equipment automatically filters out invalid segments and retains 19 seconds of valid voiceprint data as the original unauthorized voiceprint data. Information such as collection time and equipment status is recorded to ensure that the two sets of voiceprint collection equipment, sampling frequency, and verification phrase are completely consistent, eliminating interference from external factors.

[0036] In the voiceprint preprocessing stage, a standardized optimization process was performed on the two sets of original voiceprint data. First, signal denoising technology was used to remove slight equipment background noise in authorized voiceprints and slight keyboard sounds and whispers in the office area from unauthorized voiceprints, accurately preserving the core vocal characteristics of each set of voiceprints. Then, the amplitude of the two sets of denoised voiceprint signals was normalized to a unified adaptation range to eliminate amplitude deviations caused by differences in the distance between personnel and equipment and differences in vocal volume, ensuring the comparability of the two sets of signals. Finally, the two sets of normalized voiceprint signals were segmented according to a standard of 0.8s / segment, with authorized voiceprints divided into 30 segments and unauthorized voiceprints into 23 segments. The amplitude of each segment was calibrated to ensure that the amplitude fluctuation of each segment of the two sets of voiceprints did not exceed 4.5%, ultimately obtaining two sets of standardized voiceprint signals (corresponding to authorized personnel and unauthorized personnel, respectively) that meet the comparison requirements in terms of stability and consistency.

[0037] In the voiceprint visualization and extraction stage, the two sets of standardized voiceprint signals are transformed using mathematical transformation based on the Fourier transform principle. Combined with a 22kHz sampling frequency, the continuous time-domain voiceprint signals are discretized. An appropriate time window length is set, dividing the two sets of signals into several characteristically stable segments. Then, through mathematical transformation, the amplitude variation information in the time dimension is mapped to the frequency dimension, decomposing the complex time-domain signal into a superposition of sine waves with different frequencies and amplitudes. This accurately captures the fundamental frequency, three main formants, and corresponding harmonic frequency and amplitude data of each of the two sets of voiceprints, while also completely recording the phase shift of each frequency component. Based on the transformed frequency-domain quantized data, the amplitude of each key frequency point in the two sets of voiceprints is extracted. The system extracts core characteristic parameters such as peak value, mean value, gradient of change, specific frequency positions of fundamental frequency and formant, and initial phase and phase change law of each frequency component. Then, with 500-8000Hz as the effective frequency range, it generates spectrum diagrams with frequency on the horizontal axis and amplitude on the vertical axis. The energy intensity of each frequency point is intuitively presented through color gradient differences. The positions of key features such as fundamental frequency and formant are marked. Time-frequency waveform curves are generated to show the frequency change trend of the voiceprint in the time dimension. Finally, an authorized voiceprint visualization spectrum and an unauthorized voiceprint visualization spectrum containing 190 key feature points are formed. It is ensured that the frequency scale, feature point annotation standards and spectrum generation parameters of the two sets of spectra are completely consistent, laying the foundation for subsequent accurate comparison.

[0038] In the matching degree calculation stage, the feature parameters of the authorized voiceprint visualization map and the unauthorized voiceprint visualization map are precisely aligned. They are classified according to the same frequency scale, time segmentation standard and feature type to ensure that the 190 key feature points correspond one-to-one in frequency range, feature dimension and time node, forming two sets of one-to-one corresponding feature parameter sets. Then, the matching model built based on the inherent correlation characteristics of trigonometric functions first calculates each corresponding key feature point using the feature point correlation strength calculation formula, comprehensively considering the matching of the two sets of voiceprints in amplitude and phase dimensions, and obtaining the matching coefficient of each feature point. Among them, the 32 key feature points corresponding to the fundamental frequency and the three main formants play a core role in the uniqueness determination of voiceprints, and their weight is set to 1.5. The weight of the remaining 158 non-key feature points is set to 1.0. Then, the matching coefficients of all feature points are comprehensively calculated using the weighted comprehensive matching degree calculation formula, comprehensively integrating the matching differences of single feature points and the weight ratio of core features, and finally obtaining the overall matching degree value of the two sets of voiceprints of different people.

[0039] During the matching and determination phase, considering the high security requirements for access control in the core area of ​​the enterprise, it is necessary to strictly control the risk of unauthorized personnel entering. The matching threshold is set to 0.88. After the above complete process calculation, the matching degree value between the voiceprint of the unauthorized personnel and the voiceprint of the authorized personnel is 0.65. This value is far lower than the preset threshold. Therefore, it is determined that the two sets of voiceprints do not belong to the same person, that is, the voiceprint matching degree of different personnel is low, and the access control system refuses the unauthorized personnel to enter the core area. If the matching degree value is higher than or equal to 0.88, it is determined to be an authorized personnel and access is allowed. The entire implementation process clearly demonstrates the ability of this invention to identify the differences in voiceprints of different individuals through the complete matching process of voiceprints of different personnel. The matching results accurately reflect the voiceprint mismatch characteristics of unauthorized personnel and authorized personnel, effectively meeting the enterprise's internal security access control needs.

[0040] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for visual extraction and matching degree calculation of voiceprint waveforms, characterized in that, The method includes the following specific steps: Voiceprint data acquisition: Acquire the voiceprint signal of the target object through recording equipment to obtain raw voiceprint data; Voiceprint preprocessing: The original voiceprint data is denoised and normalized to remove environmental interference and redundant information, and a standardized voiceprint signal is obtained. Voiceprint visualization extraction: The standardized voiceprint signal is converted and the time-domain voiceprint signal is mapped to frequency-domain features. The amplitude, frequency and phase parameters are extracted to generate a visualized voiceprint waveform. Matching degree calculation: Extract feature parameters from two sets of visualized voiceprint waveforms, construct a matching model based on the inherent correlation characteristics of trigonometric functions, and calculate the feature fit of the two sets of voiceprints through model quantification to obtain the matching degree value; Matching determination: Set a matching threshold, and determine whether the two sets of voiceprints match based on the comparison between the calculated matching value and the threshold.

2. The method for visual extraction and matching degree calculation of voiceprint waveforms according to claim 1, characterized in that, In the voiceprint preprocessing step, signal denoising technology is used to remove environmental noise from the original voiceprint data, retain the core features of the voiceprint, normalize the amplitude of the denoised voiceprint signal to the [0,1] interval, eliminate amplitude differences caused by different acquisition devices and acquisition distances, ensure the comparability of voiceprint signals from different sources, segment the normalized voiceprint signal and perform amplitude calibration to reduce the impact of local signal fluctuations.

3. The method for visual extraction and matching degree calculation of voiceprint waveforms according to claim 1, characterized in that, In the voiceprint visualization extraction step, a mathematical transformation method based on the Fourier transform principle is used to convert the standardized voiceprint signal. The specific steps are as follows: When performing the conversion operation on the preprocessed standardized voiceprint signal using the mathematical transformation method based on the Fourier transform principle, the continuous time-domain voiceprint signal is first discretized. Combining the frequency characteristics of the voiceprint signal and the sampling frequency during acquisition, the time window length is set, and the standardized voiceprint signal is divided into several signal segments of uniform length and stable characteristics. For each signal segment, relying on the Fourier transform, the amplitude change information that fluctuates continuously with time in the time dimension is mapped to the frequency dimension, and the time-domain voiceprint signal is decomposed into a series of sine wave superpositions with different frequencies and amplitudes. Each set of sine waves corresponds to a frequency component of the voiceprint signal. During the conversion process, the amplitude and phase information corresponding to each frequency component are calculated to clarify the energy distribution characteristics of the voiceprint signal in different frequency ranges. After the conversion is completed, the obtained frequency domain signal is presented in the form of quantized frequency-amplitude and frequency-phase data pairs.

4. The method for visual extraction and matching degree calculation of voiceprint waveforms according to claim 1, characterized in that, In the voiceprint visualization extraction step, the time-domain voiceprint signal is mapped to frequency-domain features, and amplitude, frequency, and phase parameters are extracted. The specific steps are as follows: When extracting amplitude parameters, for the signal energy value corresponding to each frequency point in the frequency domain space, peak detection and effective value calculation are used to record the amplitude peak, mean, and variation gradient of different frequency intervals; when extracting frequency parameters, the effective frequency range of the voiceprint signal is focused, the fundamental frequency position and harmonic frequency interval are located through energy distribution analysis, key frequency points corresponding to resonance peaks are marked, and the distribution law of each frequency component is clarified; when extracting phase parameters, the initial phase and phase offset of each frequency component are captured, the phase change law of different frequency components during propagation is recorded, and the feature dimensions that amplitude parameters cannot cover are supplemented; after the parameters are extracted, a visualized voiceprint waveform is generated with the frequency-amplitude and frequency-phase quantized data pairs as the core.

5. The method for visual extraction and matching degree calculation of voiceprint waveforms according to claim 1, characterized in that, In the matching degree calculation step, feature parameters of two sets of visualized voiceprint waveforms are extracted. The specific steps are as follows: For the two sets of visualized voiceprint waveforms to be matched, the parameter information of the corresponding feature dimensions is matched one by one according to the same frequency scale and time segmentation standard. The amplitude quantization value and phase offset corresponding to each key frequency point are extracted, and the change trend data of each parameter in the time segment are recorded. A matching model is constructed based on the inherent correlation characteristics of trigonometric functions. The quantization value of the corresponding position in the two sets of parameter sets is used as input. The amplitude complementarity and phase coordination characteristics of sine and cosine functions are used to transform the numerical difference of the two sets of parameters into a matching degree index in the trigonometric function correlation domain. Specifically, the quantization value of each set of feature parameters is mapped to a reasonable numerical range suitable for trigonometric function operation. Then, through the inherent correspondence of trigonometric functions, the correlation strength of the two sets of parameters at the same feature point is calculated, and the feature matching coefficient of a single feature point is generated. Subsequently, the matching coefficients of all corresponding feature points are weighted and comprehensively calculated to obtain a matching degree value that can comprehensively characterize the overall matching level of the two sets of voiceprints.

6. The method for visual extraction and matching degree calculation of voiceprint waveforms according to claim 5, characterized in that, In the matching degree calculation step, the correlation strength between the two sets of parameters at the same feature point is calculated using the inherent correspondence of trigonometric functions. The calculation formula is as follows: ,in, It is the first The association strength of each feature point ranges from [0,1]. The closer the value is to 1, the higher the compatibility between the two sets of voiceprints at that feature point. The first set of voiceprints was in the... The amplitude quantization values ​​of each feature point have been normalized to [0, ] interval, The second set of voiceprints is in the... The magnitude quantization value of each feature point, and For the same frequency point, The first set of voiceprints was in the... Phase offset of each feature point The second set of voiceprints is in the... The phase offset of each feature point, and Corresponding to the same frequency point.

7. The method for visual extraction and matching degree calculation of voiceprint waveforms according to claim 6, characterized in that, In the matching degree calculation step, a weighted comprehensive calculation is performed on the matching coefficients of all corresponding feature points, and the calculation formula is as follows: ,in, This represents the final matching degree between the two sets of voiceprints, with a value ranging from [0,1]. The closer the value is to 1, the higher the overall matching degree between the two sets of voiceprints. It is the first The weight coefficients for each feature point are set according to the importance of the feature point. It is the first The correlation strength of each feature point It is the total number of feature points involved in the matching.

8. The method for visual extraction and matching degree calculation of voiceprint waveforms according to claim 1, characterized in that, In the matching determination step, the matching degree threshold is dynamically set according to the accuracy requirements of different application scenarios. For identity verification scenarios, set ≥0.9; For scenarios involving finding matching users, set ≥0.8; For scenarios that disprove biometric evidence, set ≥0.

85.

9. The method for visual extraction and matching degree calculation of voiceprint waveforms according to claim 8, characterized in that, In the matching determination step, if the calculated matching degree value ≥Preset threshold If so, then the two sets of voiceprints are determined to belong to the same object; if < If so, it is determined that the two sets of voiceprints do not belong to the same object.

Images