Caller identity authentication method and system based on dispatch call voiceprint recognition

Abstract

The invention discloses a caller identity identification method and system based on dispatch call voiceprint recognition, and the method comprises the steps: collecting voice data, achieving the preprocessing of voice through voice denoising processing, voice endpoint detection and voice speed normalization processing, carrying out the voiceprint feature calculation and extraction of the preprocessed voice data, and carrying out the recognition of the voiceprint feature of a caller. Carrying out the modeling according to the voice voiceprint features, completing the building of a voiceprint feature library, carrying out the matching of the extracted voiceprint features with the voiceprint feature library, judging the identity of a caller, and completing the identity authentication of the caller; the method can effectively prevent identity fraudulent use and fraudulent behaviors, ensures the accuracy and safety of the scheduling instruction, reduces the steps of manual verification, and improves the scheduling efficiency.

Description

Technical Field

[0001] The present invention relates to a method and system for identifying a caller's identity, and in particular to a method and system for identifying a caller's identity based on voiceprint recognition of a dispatching telephone. Background Art

[0002] In the power dispatching system, dispatching monitoring personnel need to frequently communicate with different callers, so it is particularly important to ensure the authenticity of the caller's identity. However, most current dispatching systems still rely on manual identity verification, which is verified by simple means such as asking for name and work number. This verification method has a high risk of being impersonated and affected by social engineering attacks, and may reduce the security of the dispatching system due to information leakage and other issues.

[0003] Existing identity authentication methods mainly include biometrics such as passwords, fingerprints, and face recognition. Although these methods can improve the accuracy and security of identity authentication in some situations, they also have their own shortcomings. Passwords are easily forgotten and stolen, and fingerprints and face recognition face the threat of forgery. Especially in scenarios with high security requirements such as power systems, it is particularly important to ensure the security of identity authentication.

[0004] Compared with traditional passwords or biometrics, voiceprint recognition technology has natural remote applicability, does not require physical equipment, and can improve the convenience and security of identity authentication. Therefore, applying voiceprint recognition technology to the power dispatching system can effectively improve the security and accuracy of the dispatching system in identity authentication, so as to better meet the requirements of the power dispatching system for identification technology. Summary of the invention

[0005] Purpose of the invention: The purpose of the invention is to provide a method for identifying the identity of a caller based on voiceprint recognition of a dispatch telephone to accurately identify the identity of the caller in real time and improve the security and efficiency of identity authentication. On the other hand, to provide a system for identifying the identity of a caller based on voiceprint recognition of a dispatch telephone.

[0006] Technical solution: The present invention discloses a method for identifying the identity of a caller based on voiceprint recognition of a dispatching telephone, comprising the following steps:

[0007] (1) Collecting voice data;

[0008] (2) preprocessing the collected data, wherein the preprocessing process includes denoising, speech endpoint detection, and speech rate normalization;

[0009] (3) extracting voiceprint features from the preprocessed speech data;

[0010] (4) Convert the voiceprint features into voiceprint vectors and establish a voiceprint feature library after normalization;

[0011] (5) Collecting voice data in real time, calculating the similarity between the voice to be recognized and each voiceprint vector in the feature library, and using a probabilistic linear discriminant analysis algorithm to determine whether the voice to be recognized matches a voiceprint in the feature library to complete identity authentication.

[0012] Preferably, the denoising process described in step 2 uses spectral subtraction to suppress noise on the speech data, removes interference from broadband white noise, then filters out high-frequency noise through a low-pass filter, and then uses adaptive noise suppression technology to further enhance the denoising effect, ensuring that the extracted speech features are purer.

[0013] Preferably, the speech endpoint detection in step 2 is to perform speech endpoint detection on the denoised speech data using short-time energy and short-time zero-crossing rate features to extract valid speech segments, thereby reducing the computational burden of useless data.

[0014] Preferably, the step 3 of extracting voiceprint features from the preprocessed voice data specifically comprises the following steps:

[0015] Step 31, performing time domain and frequency domain analysis;

[0016] Step 32, extracting core voiceprint features, wherein the core voiceprint features include Mel-frequency cepstral coefficients and linear prediction cepstral coefficients;

[0017] Step 33, perform dimensionality reduction and optimization of voiceprint features.

[0018] Preferably, the time domain and frequency domain analysis described in step 31 utilizes Fourier transform technology.

[0019] Preferably, the extraction of Mel-frequency cepstral coefficients in step 32 specifically includes the following steps: preprocessing the speech, fast Fourier transform, Mel filter bank, logarithmic operation, discrete cosine transform, and dynamic feature extraction.

[0020] Preferably, the extraction of linear prediction cepstral coefficients in step 32 specifically includes the following steps: using the result of linear prediction analysis to extract characteristic coefficients reflecting the shape of the vocal tract, and further supplementing the characteristic dimension of the Mel-frequency cepstral coefficients.

[0021] Preferably, the voiceprint feature dimensionality reduction in step 33 adopts principal component analysis and linear discriminant analysis dimensionality reduction methods.

[0022] Preferably, the conversion of the voiceprint features into voiceprint vectors in step 4 is performed using an x-vector voiceprint encoding algorithm.

[0023] The present invention provides a caller identity authentication system based on dispatch telephone voiceprint recognition, comprising:

[0024] Voice data acquisition module: used to receive telephone voice signals;

[0025] Preprocessing module: used to preprocess the collected voice data. The preprocessing process includes denoising, voice endpoint detection and speech rate normalization.

[0026] Voiceprint feature extraction module: used to extract voiceprint features from preprocessed voice data;

[0027] Voiceprint feature library establishment module: used to store voiceprint feature vectors, and has data encryption function, which can ensure the privacy and security of voiceprint features, and supports version control and data update;

[0028] Real-time recognition and matching judgment module: used to calculate the similarity between the voice to be recognized and each voiceprint vector in the feature library, and use the probabilistic linear discriminant analysis algorithm to judge the matching result to ensure the accuracy of identity authentication;

[0029] Identity verification module: performs identity verification based on the matching results.

[0030] Beneficial effects: Compared with the prior art, the present invention has the following advantages: 1. Real-time voiceprint recognition and identity authentication ensure the continuity and reliability of power system supply; 2. By establishing a voiceprint feature library, the security and reliability of identity authentication are improved; 3. Voiceprint recognition technology simplifies the power dispatching operation process and improves efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A schematic diagram of the process of establishing a voiceprint feature library of the present invention;

[0032] Figure 2 This is a schematic diagram of the result of real-time identification of the caller's identity by voiceprint recognition according to the present invention;

[0033] Figure 3 It is a schematic diagram of the real-time identity authentication process of the present invention. DETAILED DESCRIPTION

[0034] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings.

[0035] (1) Collecting Voice Data

[0036] A professional noise-cancelling headset was selected as the recording device, and its sampling rate was set to 16kHz to meet the requirements for the frequency resolution of audio data. Voice data was collected from daily calls or simulated scenarios of multiple operators in the power monitoring system.

[0037] During the data collection process, each monitoring personnel is required to record in simulated daily dispatch call scenarios, such as simulating the processing of power failure alarms, communicating with substation personnel in different regions, etc. During the recording process, the monitoring personnel need to show different emotional states as required, such as calmly reporting data, anxiously handling emergencies, etc. At the same time, adjust the speed and volume, from a slower and clear speed to a faster emergency reporting speed, and the volume from normal to loud to emphasize important information; record at least 20 minutes of voice data for each monitoring personnel to ensure the diversity and coverage of data samples. During the recording process, turn off unnecessary electrical equipment in the simulation room, reduce background noise sources, and use sound insulation materials to optimize the recording environment to reduce the impact of external environmental noise on voice data.

[0038] (2) Preprocessing speech data

[0039] 1. Denoising

[0040] First, spectral subtraction is used to suppress noise in the collected raw audio data to remove interference from broadband white noise such as computer fan sound and air conditioner operation sound. Then, high-frequency noise is further filtered through a low-pass filter. For example, if the sampling rate is 16000Hz and the maximum sound collection frequency is 8000Hz, the filtered sound frequency is calculated according to the formula to remove high-frequency noise components. Then, adaptive noise suppression technology is used to analyze and process the audio data in real time, and the suppression parameters are dynamically adjusted according to the changes in noise to further enhance the denoising effect and make the extracted speech features purer. The formula is:

[0041]

[0042] where f s Indicates the sampling rate of the sound device in Hz; f a is the sound collection frequency, and its maximum value is half of the sampling rate; f d Indicates the frequency of the sound after filtering.

[0043] 2. Voice endpoint detection

[0044] For the denoised audio data, its short-time energy and short-time zero-crossing rate features are calculated. By setting appropriate thresholds, the start and end points of speech can be accurately identified from long-term recordings, valid speech segments can be extracted, and silent segments in the recordings can be removed. For example, for a 5-minute recording, after speech endpoint detection, a 3-minute valid speech segment may be obtained, thereby reducing the computational burden of subsequent useless data.

[0045] 3. Normalize speech speed

[0046] The dynamic time warping (DWT) algorithm is used to standardize the speaking speed. For example, for the voice data of a monitoring personnel with a fast speaking speed, the DTW algorithm is used to match and adjust it with the standard speaking speed template, so that the rhythm of the voice data on the time axis tends to be consistent, reducing the impact of the speaking speed difference on the voiceprint feature extraction.

[0047] (3) Extracting voiceprint features

[0048] 1. Time domain and frequency domain analysis

[0049] The audio signal is converted from the time domain to the frequency domain using Fourier transform technology. For a 10-second audio signal, its frequency feature representation is obtained through fast Fourier transform, and the energy distribution of different frequency components is analyzed to provide basic data for subsequent core feature extraction.

[0050] 2. Core feature extraction

[0051] Mel-frequency cepstral coefficient (MFCC) extraction:

[0052] The audio is pre-emphasized to increase the energy of the high-frequency part, and then frame-by-frame windowing is performed, with each frame length set to 25ms and a frame shift of 10ms. Each frame of audio data is subjected to a fast Fourier transform to obtain a spectrum. The spectrum is passed through a Mel filter group, which is evenly distributed on the Mel frequency scale, with a total of 26 filters set to obtain a Mel spectrum. The Mel spectrum is logarithmically operated, and then a discrete cosine transform is performed, and the first 13 Mel-frequency cepstral coefficients are taken as features. Finally, dynamic feature extraction is performed, and the first and second order differences of the MFCC features are calculated to obtain a Mel-frequency cepstral coefficient feature vector containing static and dynamic features.

[0053] Linear Prediction Cepstral Coefficient (LPCC) Extraction:

[0054] The audio data is modeled using linear prediction analysis (LPC) to calculate prediction coefficients. For example, the 16th-order linear prediction cepstral coefficients are calculated using the autocorrelation method; the linear prediction cepstral coefficient features are calculated based on the LPC coefficients, and the feature dimensions of the Mel-frequency cepstral coefficients are further supplemented to obtain a more comprehensive voiceprint feature representation.

[0055] 3. Feature Dimensionality Reduction and Optimization

[0056] The principal component analysis is used to reduce the dimension of the extracted MFCC and LPCC feature vectors. For example, the original feature vector dimension is 39 (13 MFCC coefficients and their first-order and second-order differences have a total of 39 dimensions). The PCA algorithm is used to reduce the dimension to 20 dimensions, remove redundant information, reduce the data dimension, and retain the feature discrimination as much as possible. Then, the linear discriminant analysis is used to further optimize the feature vector and improve the distinguishability between the voiceprint features of different monitored personnel.

[0057] (4) Establish a voiceprint feature library

[0058] A relational database is used to store voiceprint feature vectors. An independent table is created for each power monitoring personnel. The table contains voiceprint feature vector fields and corresponding personnel identity information fields, such as work number, name, etc. The work number is used as the primary key for index management to facilitate quick search and similarity calculation. Data encryption function is enabled at the database level, such as using the AES encryption algorithm to encrypt and store voiceprint feature vectors to ensure the privacy and security of voiceprint features.

[0059] A version number field is established for the voiceprint feature library of each monitoring personnel. When the data is updated or the voiceprint data is re-collected, the version number is updated, and the timestamp of each update and the collection environment parameters, such as temperature, humidity, recording device model, etc., are recorded. For example, the initial version of the voiceprint feature library of monitoring personnel A is 1.0. After an equipment upgrade, the voiceprint data is re-collected. The updated version is 1.1, and the update time is recorded as 12:00 on December 10, 2024. The collection environment temperature is 25℃, the humidity is 60%, and a professional headset is selected as the recording device.

[0060] (5) Voiceprint matching and identity verification

[0061] During the actual dispatch call process, voice data is collected in real time through the voice collection module connected to the telephone system. For example, when a new dispatch call is connected, the system automatically starts voice collection and collects a segment of voice data every 5 seconds; the collected voice data to be recognized is subjected to the same preprocessing and feature extraction steps as before to obtain the voiceprint feature vector of the voice to be recognized; the similarity between the voiceprint feature vector of the voice to be recognized and each voiceprint vector in the voiceprint feature library is calculated using the probabilistic linear discriminant analysis similarity algorithm. For example, in the experiment, the similarity threshold is set to 0.95. If the similarity between the voice to be recognized and the voiceprint feature vector of a certain monitoring person is greater than 0.95, the voice is judged to match the monitoring person and the identity authentication is passed; otherwise, it is judged that the identity authentication fails, and an alarm is issued to indicate that there may be an abnormality, ensuring the accuracy and security of identity authentication.

Claims

1. A method for identifying the identity of a caller based on voiceprint recognition of a dispatching telephone, characterized in that: The following steps are involved: (1) Collecting voice data; (2) preprocessing the collected data, wherein the preprocessing process includes denoising, speech endpoint detection, and speech rate normalization; (3) extracting voiceprint features from the preprocessed speech data; (4) Convert the voiceprint features into voiceprint vectors and establish a voiceprint feature library after normalization; (5) Collecting voice data in real time, calculating the similarity between the voice to be recognized and each voiceprint vector in the feature library, and using a probabilistic linear discriminant analysis algorithm to determine whether the voice to be recognized matches a voiceprint in the feature library to complete identity authentication.

2. The method for identifying the identity of a caller based on voiceprint recognition of a dispatching telephone according to claim 1, characterized in that: The denoising process described in step 2 uses spectral subtraction to suppress noise on the speech data, then filters out high-frequency noise through a low-pass filter, and then uses adaptive noise suppression technology to further enhance the denoising effect.

3. The method for identifying the identity of a caller based on voiceprint recognition of a dispatching telephone according to claim 1, characterized in that: The speech endpoint detection described in step 2 is to perform speech endpoint detection on the denoised speech data using short-time energy and short-time zero-crossing rate features to extract valid speech segments.

4. The method for identifying the identity of a caller based on voiceprint recognition of a dispatching telephone according to claim 1, characterized in that: Step 3 extracts voiceprint features from the preprocessed voice data, and the specific steps include: Step 31, performing time domain and frequency domain analysis; Step 32, extracting core voiceprint features, wherein the core voiceprint features include Mel-frequency cepstral coefficients and linear prediction cepstral coefficients; Step 33, perform dimensionality reduction and optimization of voiceprint features.

5. The method for identifying the identity of a caller based on voiceprint recognition of a dispatching telephone according to claim 4, characterized in that: The time domain and frequency domain analysis described in step 31 utilizes Fourier transform technology.

6. The method for identifying the identity of a caller based on voiceprint recognition of a dispatching telephone according to claim 4, characterized in that: The extraction of Mel-frequency cepstral coefficients in step 32 specifically includes the following steps: preprocessing the speech, fast Fourier transform, Mel filter bank, logarithmic operation, discrete cosine transform, and dynamic feature extraction.

7. The method for identifying the identity of a caller based on voiceprint recognition of a dispatching telephone according to claim 4, characterized in that: The extraction of linear prediction cepstral coefficients in step 32 specifically includes the following steps: using the result of linear prediction analysis to extract characteristic coefficients reflecting the shape of the vocal tract, and further supplementing the characteristic dimension of the Mel-frequency cepstral coefficients.

8. The method for identifying the identity of a caller based on voiceprint recognition of a dispatching telephone according to claim 4, characterized in that: The voiceprint feature dimensionality reduction described in step 33 adopts principal component analysis and linear discriminant analysis dimensionality reduction methods.

9. The method for identifying the identity of a caller based on voiceprint recognition of a dispatching telephone according to claim 1, characterized in that: In step 4, the voiceprint features are converted into voiceprint vectors using the x-vector voiceprint encoding algorithm.

10. A caller identification system based on dispatch telephone voiceprint recognition, characterized in that: include: Voice data acquisition module: used to receive telephone voice signals; Preprocessing module: used to preprocess the collected voice data. The preprocessing process includes denoising, voice endpoint detection and speech rate normalization. Voiceprint feature extraction module: used to extract voiceprint features from preprocessed voice data; Voiceprint feature library building module: used to store voiceprint feature vectors; Real-time recognition and matching judgment module: used to calculate the similarity between the speech to be recognized and each voiceprint vector in the feature library, and use the probabilistic linear discriminant analysis algorithm to determine the matching result; Identity authentication module: performs identity authentication based on the matching results.

Images