CN-116665679-B - Intelligent voiceprint anti-counterfeiting method, device and equipment based on hops signals

CN116665679BCN 116665679 BCN116665679 BCN 116665679BCN-116665679-B

Abstract

The invention discloses an intelligent voiceprint anti-counterfeiting method, device and equipment based on a hops signal, which are characterized in that voiceprint signals generated by striking hops on the bottle wall are collected, then PLP characteristics are extracted after denoising treatment is carried out on the voiceprint signals, the characteristic signals are input into a CNN-GMM-HMM voiceprint model for training to construct an original standard database, after the sample voiceprint signals are collected, the sample voiceprint signals are preprocessed and then are matched and identified with voiceprint information in the standard database, so that an authenticity verification effect is achieved.

Inventors

LIU GUODONG
ZHANG ZHUOQING
LI ZHIJIAN
WU JIAHAO
MENG QINGJUN

Assignees

陕西科技大学

Dates

Publication Date: 20260505
Application Date: 20230625

Claims (7)

1. An intelligent voiceprint anti-counterfeiting method based on hop signals is characterized by comprising the following steps of: step 1, collecting a plurality of original voiceprint signals; step 2, denoising pre-characterization processing is carried out on the collected voiceprint signals through a self-encoder; The self-encoder is a stacked sparse self-encoder, inputs the collected original voiceprint signals into a deep neural network model SAE formed by a plurality of layers of sparse self-encoders for training, then removes a decoding layer, takes the signals processed by the first training characteristics as the input of the second SAE training, generates second characteristics, and then decodes and outputs the second characteristics to obtain data after pre-characterization processing; Step 3, extracting PLP features from the pre-characterized data in the step 2 by using a voice feature extraction tool, and then inputting the PLP features into a CNN-GMM-HMM voiceprint model for learning training to obtain a CNN-GMM-HMM voiceprint model standard database; The CNN-GMM-HMM voiceprint model comprises a CNN module, a GMM module, an HMM module, a training module and a training module, wherein the CNN module is used for extracting and classifying PLP features through a convolution layer, a pooling layer and a full connection layer; Step 4, repeatedly collecting the white wine to be detected through the acoustic vibration sensor for a plurality of times to obtain a sample hops voiceprint signal; step 5, carrying out the same denoising pre-characterization treatment on the collected sample hops voiceprint signals through the stacking sparse self-encoder in the step 2; Step 6, inputting the voiceprint signals preprocessed in the step 5 into a standard database in the step3 for identification comparison; and 7, displaying and outputting the identification verification result.
2. The intelligent voiceprint anti-counterfeiting method based on the hop signals is characterized in that in the step 3, PLP features are extracted by using a voice feature extraction tool OpenSMILE after the pre-characterization processing in the step 2, firstly, frame windowing processing is carried out on voiceprint data, voiceprint signal fragments are acquired, then, fast Fourier transformation is carried out to obtain a frequency spectrum, then, amplitude square operation is carried out to carry out amplification processing, then, bark filter bank processing is carried out, then, equal loudness pre-emphasis and intensity-loudness conversion are carried out, and then, inverse Fourier transformation is carried out to carry out linear prediction to obtain the PLP features.
3. The intelligent voiceprint anti-counterfeiting method based on hop signals, as set forth in claim 1, is characterized in that in the step 6, PLP features are in a voiceprint model, a separate convolution layer is used as input, independent operation is performed on each channel, maximum pooling is performed across channels, the channel with the largest response in each node is selected, PLP features are extracted and classified, then the extracted PLP features are used for cluster training in a GMM model, and phonemes are represented as hidden variables to be imported into the HMM model for training; the GMM-HMM training is started, the GMM model is initialized by adopting an average distribution mode, and then the maximum likelihood estimated value of the hidden variable is calculated by utilizing the existing estimated value of the hidden variable, namely the average value of the sample Sum of variances 2 : 2 And then realignment is carried out, the voiceprint signals are aligned according to the obtained transition probability, the mean value and the variance, after basic parameters are set, repeated training times are set for training until convergence, and then the CNN-GMM-HMM voiceprint model standard database is obtained.
4. The intelligent voiceprint anti-counterfeiting method based on hop signals according to claim 1, wherein the identification comparison in the step 6 is specifically: firstly, calculating the error acceptance rate and the error rejection rate corresponding to each threshold point of the voiceprint signal preprocessed in the step 5; then drawing an ROC curve according to the calculated accuracy and recall rate of each threshold point; And finally, searching an abscissa X value corresponding to the intersection point of the ROC curve and the diagonal line in the ROC space, wherein the smaller the value is, the higher the true degree of the detected white spirit sample is.
5. Intelligent voiceprint anti-counterfeiting device based on hops signals is characterized by comprising: The hops voiceprint signal excitation unit is used for repeatedly shaking the test sample by utilizing attractive/repulsive magnetic force through a circularly alternating magnetic field generated by electromagnetic induction, so as to generate hops voiceprint signals; The hops voiceprint signal acquisition unit is an acoustic vibration sensor formed by piezoelectric ceramics and damping materials and is used for detecting the vibration generated by hops in the wine bottle and the local inside of the bottle wall to obtain sample hops voiceprint signals; a hops voiceprint signal identification unit configured to perform the intelligent voiceprint anti-counterfeiting method based on hops signals according to any one of claims 1-4, and identify the voiceprint characteristics acquired by the hops voiceprint signal acquisition unit with a standard database to obtain a verification result of the hops voiceprint signals.
6. The intelligent voiceprint anti-counterfeiting device based on a hop signal according to claim 5, wherein the hop voiceprint signal excitation unit comprises a first braking frame (1) and a second braking frame (2); The first braking frame (1) is used as an integral support and consists of a horizontal moving seat (11) and a vertical driving frame (12), the vertical driving frame (12) is installed on the horizontal moving seat (11), the second braking frame (2) is hung on the vertical driving frame (12), the second braking frame (2) for placing a white wine sample consists of a white wine bottle neck clamping groove (21), a coil (22) capable of being electrified with alternating current and a white wine bottle supporting groove (23), the coil (22) capable of being electrified with alternating current is arranged at one end of the white wine bottle supporting groove (23), the white wine bottle neck clamping groove (21) is arranged at the other end of the white wine bottle supporting groove (23), the magnet (13) is arranged on one side, close to the coil (22) capable of being electrified with alternating current, of the hops voiceprint signal acquisition unit, and the sound vibration sensor is placed at the center of the white wine bottle neck clamping groove (21).
7. The intelligent voiceprint anti-counterfeiting equipment based on the hops signals is characterized by comprising a cloud server ECS and a file storage NAS; the cloud server ECS for executing a program of the intelligent voiceprint anti-counterfeiting method based on a hop signal according to any one of claims 1 to 4; the file storage NAS is used for storing a computer program and a constructed CNN-GMM-HMM voiceprint model standard database, and the computer program realizes the method according to any one of claims 1-4 when being executed by a processor.

Description

Intelligent voiceprint anti-counterfeiting method, device and equipment based on hops signals Technical Field The invention belongs to the technical field of anti-counterfeiting, relates to digital informationized anti-counterfeiting technology, and particularly relates to an intelligent voiceprint anti-counterfeiting method, device and equipment based on hops signals. Background With the continuous improvement of living standard, the market scale of white spirit gradually increases, and meanwhile, the phenomenon of fake white spirit is also increased. In order to better distinguish the authenticity of the white spirit, the anti-counterfeiting technology of the white spirit is widely applied. At present, the main stream mode of white spirit anti-counterfeiting is mainly based on physical anti-counterfeiting means of packaging, such as a label, a bottle cap, an identification code and the like, and the problems of easy imitation, difficult identification and the like exist, so that the circulation of counterfeit and inferior products in the market is caused. The information technology anti-counterfeiting mode which is an anti-counterfeiting mode by means of computer technology, information security technology and the like has the advantages of being strong in concealment, expandability, visual and the like. At present, digital information technology such as digital watermark, encryption technology and two-dimensional code is used for manufacturing anti-counterfeiting marks and patterns to improve anti-counterfeiting performance of white spirit packaging, but the digital information technology anti-counterfeiting mode has the defect that the digital watermark, the encryption technology or the two-dimensional code is easy to attack, crack or tamper generally. Disclosure of Invention The invention aims to overcome the defects of the prior art, and provides an intelligent voiceprint anti-counterfeiting method, device and equipment based on hops signals, which are used for distinguishing true and false of white wine based on unique voiceprint signals of white wine hops, and have the advantages of being safer, more reliable and more efficient. In order to achieve the above object, the present invention adopts the following technical scheme. An intelligent voiceprint anti-counterfeiting method based on hop signals comprises the following steps: step 1, collecting a plurality of original voiceprint signals; step 2, denoising pre-characterization processing is carried out on the collected voiceprint signals through a self-encoder; The self-encoder is a stacked sparse self-encoder, inputs the collected original voiceprint signals into a deep neural network model SAE formed by a plurality of layers of sparse self-encoders for training, then removes a decoding layer, takes the signals processed by the first training characteristics as the input of the second SAE training, generates second characteristics, and then decodes and outputs the second characteristics to obtain data after pre-characterization processing; Step 3, extracting PLP features from the pre-characterized data in the step 2 by using a voice feature extraction tool, and then inputting the PLP features into a CNN-GMM-HMM voiceprint model for learning training to obtain a CNN-GMM-HMM voiceprint model standard database; The CNN-GMM-HMM voiceprint model comprises a CNN module, a GMM module, an HMM module, a training module and a training module, wherein the CNN module is used for extracting and classifying PLP features through a convolution layer, a pooling layer and a full connection layer; Step 4, repeatedly collecting the white wine to be detected through the acoustic vibration sensor for a plurality of times to obtain a sample hops voiceprint signal; step 5, carrying out the same denoising pre-characterization treatment on the collected sample hops voiceprint signals through the stacking sparse self-encoder in the step 2; Step 6, inputting the voiceprint signals preprocessed in the step 5 into a standard database in the step3 for identification comparison; and 7, displaying and outputting the identification verification result. In step 3, the PLP features are extracted by using the speech feature extraction tool OpenSMILE after the pre-characterization processing in step2, firstly, the frame-division windowing processing is performed on the voiceprint data, the acquired voiceprint signal fragments are intercepted, then the fast fourier transform is performed to obtain a frequency spectrum, then the amplitude square operation is performed to amplify the frequency spectrum, then the Bark filter bank processing is performed, then the equal loudness pre-emphasis and the intensity-loudness conversion are performed, and then the inverse fourier transform is performed to perform the linear prediction to obtain the PLP features. In step 6, the PLP features are in a voiceprint model, a single convolution layer is used as input, the amplitude spectrum is independen