Method for constructing sound classification model, sound classification method and system

Abstract

The present invention relates to a method for constructing a sound classification model, a sound classification method and a system, including preprocessing, preprocessing original sound data; feature extraction, pre-emphasis, framing, windowing, fast Fourier transform, Mel filter bank and discrete cosine transform of sound signals to extract spectrum features; modeling, using extreme learning machine algorithm to randomly generate connection weights of input layer and hidden layer and thresholds of neurons in hidden layer, training the extracted feature data to obtain a training database, and performing empirical analysis and parameter optimization through extreme learning machine classifier to obtain a classification model. The advantages of the present invention are: using Mel cepstral coefficient algorithm to extract natural environment sound, then using extreme learning machine as classifier to train and test the extracted audio features, and finally analyzing the feature category to which the sound to be tested belongs, thereby realizing the recognition and classification of sound, which has obvious advantages.

Description

Technical Field

[0001] The present invention relates to the field of sound recognition technology, and in particular to a method for constructing a sound classification model, a sound classification method and a system. Background Art

[0002] Sound contains a wealth of information and is one of the components of the human sensory world. Environmental audio data is closely related to the lives of animals and humans. For example, the use of automatic monitoring and identification of species sounds in forests and wilderness can be used to scout endangered animals; the use of automatic monitoring and identification of abnormal animal sounds in earthquake zones can be used as an auxiliary technology for earthquake early warning systems. It can be seen that the recognition and classification of natural environmental sounds are closely related to humans. Therefore, we can help understand biodiversity and monitor the balance of the ecosystem by extracting and analyzing the information characteristics contained in natural environmental sounds. By analyzing various environmental sounds, we can obtain environmental information, better protect the environment, and achieve development between man and nature.

[0003] From the perspective of environmental sound recognition technology, it involves the intersection of multiple disciplines, such as digital signal processing, auditory mechanism, pattern recognition, machine learning, etc. Environmental sound recognition is used in many fields, such as audio forensics, sound scene analysis, environmental safety monitoring, positioning tracking and sound source classification, patient monitoring and abnormal event detection, etc. However, the current research on environmental sound is relatively small and a mature recognition method has not yet been formed. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method for constructing a sound classification model based on an extreme learning machine for feature extraction and recognition, and a sound classification method and system.

[0005] The technical solution adopted by the present invention to solve the technical problem is: a method for constructing a sound classification model, comprising:

[0006] Preprocessing: preprocessing the original sound data;

[0007] Feature extraction: pre-emphasis, framing, windowing, fast Fourier transform, Mel filter bank and discrete cosine transform are performed on the sound signal to extract spectral features;

[0008] Modeling, using the extreme learning machine algorithm to randomly generate the connection weights of the input layer and the hidden layer and the thresholds of the hidden layer neurons, the extracted feature data are trained to obtain a training database, and the extreme learning machine classifier is used for empirical analysis and parameter optimization to obtain a classification model.

[0009] Preferably, the preprocessing includes converting the original sound data into a wav format, using a filter to eliminate environmental noise for each sound sample, cutting out the noise and silence parts in the sound waveform, and retaining the characteristic waveform.

[0010] Preferably, the method for pre-emphasis of the sound signal is to use a high-pass filter to enhance the high-frequency part of the sound signal, and the high-pass filter is,

[0011] y(n)=x(n)-α*x(n-1) (1)

[0012] Wherein, 0.9≤α≤1.0, n is the number of sampling points of the signal, x(n) is the sound signal, and y(n) is the sound signal after pre-emphasis.

[0013] Preferably, the frame division method is to divide the sound signal after pre-emphasis processing into multiple frames with a duration of t.

[0014] Preferably, the windowing method is:

[0015] The signal after framing is expressed as S(n), n=0,1,…N-1, where n is the number of frames obtained after framing, and N is the size of the frame. After adding the Hamming window, it is expressed as S′(n)=S(n)*W(n). The form of the Hamming window W(n) is,

[0016]

[0017] Among them, a=0.46.

[0018] Preferably, the method for performing fast Fourier transform on the sound signal is to create a Hamming window matrix C, whose size is the same as S(n), and multiply the corresponding positions of the two matrices to obtain a windowed sound information matrix S ′ (n), perform FFT fast Fourier transform on the windowed sound information matrix to obtain matrix D, take the modulus and square the data points of each frame in matrix D, calculate the energy spectrum density E, add the energy obtained for each frame to obtain a new matrix F, in which each element represents the sum of the energy of each frame.

[0019] Preferably, the method of performing Mel filter bank and discrete cosine transform to extract spectral features is:

[0020] Define a filter bank with M filters. The filter used is a triangular filter, and its frequency response is defined as:

[0021]

[0022] in, Spectral line index k = (1 + L).f(m) / f s, L is the number of FFT points, f(m) is the value after the calculated Mel scale is converted to frequency, f s is the sampling frequency;

[0023] The logarithmic energy of each filter bank output is then calculated as,

[0024]

[0025] Among them, s(m) is the logarithmic energy, X(k) is the energy spectrum of the sound signal, and H m (k) is a filter bank;

[0026] After the energy of different frequency bands is logarithmically transformed, it is transformed back to the time domain through discrete cosine transform, which is called Mel-frequency cepstrum parameter (MFCC). Its calculation formula is as follows:

[0027]

[0028] Preferably, the modeling process includes two stages: random feature mapping and linear parameter solving;

[0029] In the first stage, the hidden layer parameters are randomly initialized, and some nonlinear mappings are used as activation functions to map the input data to a new feature space. The output of the hidden layer is denoted as H(x), and the calculation formula is:

[0030] H(x)=[h1(x),…,h L (x)] (6)

[0031] h i (x) is represented as follows:

[0032] h i (x) = g(w i ,b i ,x)=g(w i x+b i ),w i ∈R D ,b i ∈R (7)

[0033] where w i and b i is the hidden layer node parameter, g(w i ,v i ,x) is the activation function, and the hidden layer output H is calculated according to formulas (6) and (7);

[0034] In the second stage, the weight (β) connecting the hidden layer and the output layer is solved by minimizing the approximate square error. The objective function is as follows:

[0035] min‖Hβ-T‖2 ,β∈R L*m (8)

[0036] Where H is the output matrix of the hidden layer and T is the target matrix of the training data:

[0037]

[0038] Through the knowledge of linear algebra and matrix theory, the optimal solution of formula (8) can be derived as:

[0039] β * =H + T (10)

[0040] Among them, H + is the Moore–Penrose generalized inverse matrix of matrix H, H + =(H T H) -1 H T , the training is now completed.

[0041] The present invention also provides a sound classification method based on an extreme learning machine, comprising:

[0042] An extreme learning machine classification model is obtained by training based on the model building method;

[0043] The sound signal to be classified is input into the extreme learning machine classification model to obtain the classification result.

[0044] The present invention also provides a sound classification system based on the construction method of the sound classification model, comprising:

[0045] The sound pickup includes a microphone and an audio amplifier circuit, and is used for collecting sound signals.

[0046] A main processing module, including a control chip embedded with a preprocessing algorithm, wherein the preprocessing algorithm is used to receive and preprocess the sound signal obtained by the sound pickup;

[0047] A DSP algorithm processing module is communicatively connected with the main processing module, and includes a chip implanted with a recognition algorithm, wherein the recognition algorithm is the classification model obtained through training, and realizes recognition and classification of natural environment sounds;

[0048] The display module includes a display and a drive controller, which is used to display the execution status of the process of collecting and processing sound signals, including system startup, signal processing, and recognition results;

[0049] The keyboard module starts and shuts down the system through the keyboard, and controls the execution of the system through different keys on the keyboard, including displaying the steps in progress and pausing the current system operation;

[0050] A power supply and crystal oscillator module, wherein the power supply voltage is 2V to 3.6V; and the crystal oscillator module is used to provide a clock signal.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] The Mel-cephalometric coefficient algorithm is used to extract natural environment sounds, and then the extreme learning machine is used as a classifier to train and test the extracted audio features. Finally, the feature category of the sound to be tested is analyzed to achieve sound recognition and classification. It can not only effectively identify and classify different sounds, but also has obvious advantages in determining and applying analysis methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings in the following descriptions are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without creative work. By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, purposes and advantages of the present invention will become more obvious:

[0054] Figure 1 is a flow chart of feature extraction in the method for constructing a sound classification model provided by an embodiment of the present invention;

[0055] Figure 2 is a flow chart of an extreme learning classification method in a method for constructing a sound classification model provided by an embodiment of the present invention;

[0056] Figure 3 is a schematic diagram of a sound classification system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several variations and improvements may be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0058] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein, for example, can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0059] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0060] Example 1

[0061] Combination Figure 1 and Figure 2 , this embodiment provides a sound classification method, comprising the following steps:

[0062] Preprocessing: preprocessing the original sound data;

[0063] Feature extraction: pre-emphasis, framing, windowing, fast Fourier transform, Mel filter bank and discrete cosine transform are performed on the sound signal to extract spectral features;

[0064] Modeling, using the extreme learning machine algorithm to randomly generate the connection weights of the input layer and the hidden layer and the thresholds of the hidden layer neurons, the extracted feature data are trained to obtain a training database, and the extreme learning machine classifier is used for empirical analysis and parameter optimization to obtain a classification model.

[0065] This embodiment uses the Mel-frequency cepstral coefficient algorithm to extract natural environment sounds, and then uses the extreme learning machine as a classifier to train and test the extracted audio features. Finally, the feature category to which the sound to be tested belongs is analyzed, thereby realizing the recognition and classification of the sound. It can not only effectively identify and classify different sounds, but also has obvious advantages in determining and applying the analysis method.

[0066] Specifically, the method for constructing a sound classification model provided in this embodiment includes:

[0067] Preprocessing: preprocessing the original sound data;

[0068] The preprocessing includes converting the original sound data into wav format, using a filter to eliminate environmental noise for each sound sample, cutting out the noise and silence parts in the sound waveform, and retaining the characteristic waveform, thereby reducing the signal length and the amount of calculation.

[0069] Feature extraction: use the Mel-frequency cepstral coefficient (CMCC) algorithm to pre-emphasize, frame, window, fast Fourier transform, Mel filter bank and discrete cosine transform to extract spectral features of the sound signal;

[0070] The Mel-scale Cepstral Coefficient (MFCC) algorithm can be used to extract the characteristics of sound signals. The Mel scale describes the nonlinear characteristics of human ear frequency. Its relationship with frequency can be approximately expressed by the following formula:

[0071]

[0072] The method for pre-emphasis of the sound signal is to use a high-pass filter to enhance the high-frequency part of the sound signal, wherein the high-pass filter is:

[0073] y(n)=x(n)-α*x(n-1) (2)

[0074] Wherein, 0.9≤α≤1.0, in this embodiment, 0.97 is taken, n is the number of sampling points of the signal, x(n) is the sound signal, and y(n) is the sound signal after pre-emphasis.

[0075] The framing method is to divide the pre-emphasized sound signal into multiple frames of duration t, and retain part of the time domain information by framing before converting the time domain information into the frequency domain through Fourier transform, so as to properly express the frequency domain and time domain information of the speech signal.

[0076] The sampling frequency of the speech signal used in speech recognition is 8KHz or 16KHz, so the frame interval t is adaptively taken as 256 or 512.

[0077] The windowing method is to add each frame of data to a Hamming window of size 1103, and set the value outside the window to 0. The signal after framing is expressed as S(n), n = 0, 1, ... N-1, where n is the number of frames obtained after framing, and N is the size of the frame, that is, the number of sampling points that cross between frames. After adding the Hamming window, it is expressed as S ′ (n) = S(n)*W(n), then the form of W(n) is,

[0078]

[0079] Different a values ​​will produce different Hamming windows, and generally a=0.46.

[0080] After multiplying the Hamming window, each frame must be subjected to a fast Fourier transform to obtain the energy distribution on the spectrum, and the spectrum of the speech signal is squared to obtain the power spectrum of the speech signal. The specific method is to create a Hamming window matrix C, which is the same size as S(n), and multiply the corresponding positions of the two matrices to obtain the windowed sound information matrix S ′ (n), perform FFT fast Fourier transform on the windowed sound information matrix to obtain matrix D, take the modulus and square the data points of each frame in matrix D, calculate the energy spectrum density E, add the energy obtained for each frame, and obtain a new matrix F, in which each element represents the sum of the energy of each frame.

[0081] Then define a filter bank with M filters. The filter used is a triangular filter, and its frequency response is defined as:

[0082]

[0083] in,

[0084] The logarithmic energy of each filter bank output is calculated as,

[0085]

[0086] Among them, s(m) is the logarithmic energy, X(k) is the energy spectrum of the sound signal, and H m (k) is a filter bank;

[0087] After the energy of different frequency bands is logarithmically transformed, it is transformed back to the time domain through discrete cosine transform, which is called Mel-frequency cepstrum parameter (MFCC). Its calculation formula is as follows:

[0088]

[0089] Modeling: using the extreme learning machine algorithm to randomly generate the connection weights of the input layer and the hidden layer and the thresholds of the neurons in the hidden layer, training the extracted feature data to obtain a training database, and performing empirical analysis and parameter optimization through the extreme learning machine classifier to obtain a classification model;

[0090] The modeling process includes two stages: random feature mapping and linear parameter solving;

[0091] In the first stage, the hidden layer parameters are randomly initialized, and some nonlinear mappings are used as activation functions to map the input data to a new feature space. The output of the hidden layer is denoted as H(x), and the calculation formula is:

[0092] H(x)=[h1(x),…,h L (x)] (7)

[0093] h i (x) is represented as follows:

[0094] h i (x) = f(w i ,b i ,x)=g(w i x+b i ),w i ∈R D ,b i ∈R (8)

[0095] where w i and b i is the hidden layer node parameter, g(w i ,b i ,x) is the activation function, and the hidden layer output H is calculated according to formulas (7) and (8);

[0096] In the second stage, the weight (β) connecting the hidden layer and the output layer is solved by minimizing the approximate square error. The objective function is as follows:

[0097] min‖Hβ-T‖ 2 ,β∈R L*m (9)

[0098] Where H is the output matrix of the hidden layer and T is the target matrix of the training data:

[0099]

[0100] Through the knowledge of linear algebra and matrix theory, the optimal solution of formula (9) can be derived as:

[0101] β * =H + T (11)

[0102] Among them, H + is the Moore–Penrose generalized inverse matrix of matrix H, H + =(H T H) -1 H T , the training is now completed. During testing, the training results can be used to predict the results, and the extreme learning machine model can be used to realize the recognition of natural environment sounds.

[0103] Example 2

[0104] This embodiment discloses a sound classification method, including:

[0105] An extreme learning machine classification model is obtained by training based on the sound classification model construction method provided in Example 1;

[0106] The sound signal to be classified is input into the extreme learning machine classification model to obtain the classification result.

[0107] Example 3

[0108] refer to Figure 3 , this embodiment provides a sound classification system, including,

[0109] The sound pickup includes a microphone and an audio amplifier circuit, and is used for collecting sound signals.

[0110] A main processing module, including a control chip embedded with a preprocessing algorithm, wherein the preprocessing algorithm is used to receive and preprocess the sound signal obtained by the sound pickup;

[0111] The control chip is STM32F103T8, which has 64KB flash memory and 20KB running memory. The package size is small, the maximum operating voltage is 3.6V, and a low-dropout linear regulator is used to reduce the input voltage to 3.3V to power the microcontroller;

[0112] A DSP algorithm processing module is communicatively connected with the main processing module, and includes a chip implanted with a recognition algorithm, wherein the recognition algorithm is the classification model obtained through training, and realizes recognition and classification of natural environment sounds;

[0113] The DSP algorithm processing module uses a chip model of TMS320VC5509A, which is a fixed-point audio-specific DSP processor based on the TI C55x architecture. It improves parallelism while reducing energy dissipation, achieving high performance and low power consumption. The DSP module is used to store the feature-extracted data in the flash memory, and then the data is trained using the extreme learning machine learning method, and then the data is processed to achieve natural environment sound recognition and classification;

[0114] The display module includes a display and a drive controller, which is used to display the execution status of the process of collecting and processing sound signals, including system startup, signal processing, and recognition results;

[0115] The display module adopts an OLCD display screen and a dedicated liquid crystal drive controller with a chip model of SVB-LT0701;

[0116] A keyboard module is used to start and shut down the system through the keyboard, and to control the execution of the system through different keys on the keyboard, including displaying the ongoing steps and pausing the current system operation; this embodiment uses a 4×4 keyboard module;

[0117] A power supply and crystal oscillator module, wherein the power supply voltage is 2V to 3.6V; the crystal oscillator module is used to provide a clock signal;

[0118] The power supply voltage of STM32F103C8T6 is 2V~3.6V. You only need to design an XH socket in the circuit and use a 3.7V lithium battery for power supply. The access voltage is not allowed to exceed 6V. The crystal oscillator circuit is used to provide a clock signal to the chip. Without a crystal oscillator, the microcontroller cannot work. The larger the crystal oscillator value, the faster the microcontroller runs. However, the faster the speed, the easier it is to be interfered with, and the worse the reliability. Therefore, you can choose a suitable power supply and crystal oscillator according to the design requirements.

[0119] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices (systems) or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0120] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the embodiments can be referred to each other. The above description of the disclosed embodiments enables professionals and technicians in this field to implement or use the present invention. Various modifications to these embodiments will be obvious to professionals and technicians in this field, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown in this article, but will comply with the widest range consistent with the principles and novel features disclosed herein.

[0121] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art may make various modifications or variations within the scope of the claims, which do not affect the essence of the present invention.

Claims

1. A method for constructing a sound classification model, characterized in that: include Preprocessing: preprocessing the original sound data; Feature extraction: pre-emphasis, framing, windowing, fast Fourier transform, Mel filter bank and discrete cosine transform are performed on the sound signal to extract spectral features; Modeling: using the extreme learning machine algorithm to randomly generate the connection weights of the input layer and the hidden layer and the thresholds of the neurons in the hidden layer, training the extracted feature data to obtain a training database, and performing empirical analysis and parameter optimization through the extreme learning machine classifier to obtain a classification model; The modeling process includes two stages: random feature mapping and linear parameter solving; In the first stage, the hidden layer parameters are randomly initialized, and some nonlinear mappings are used as activation functions to map the input data to a new feature space. The output of the hidden layer is denoted as H(x), and the calculation formula is: H(x)=[h1(x),…,h L (x)] (6) h i (x) is represented as follows: h i (x)=g(w i ,b i ,x)=g(w i x+b i ),w i ∈R D ,b i ∈R (7) where w i and b i is the hidden layer node parameter, g(w i ,b i ,x) is the activation function, and the hidden layer output H is calculated according to formulas (6) and (7); In the second stage, the weight (β) connecting the hidden layer and the output layer is solved by minimizing the approximate square error. The objective function is as follows: min‖Hβ-T‖ 2 ,β∈R L*m (8) Where H is the output matrix of the hidden layer, T is the target matrix of the training data, Through the knowledge of linear algebra and matrix theory, the optimal solution of formula (8) can be deduced as: b * =H + T (10) Among them, H + is the Moore–Penrose generalized inverse matrix of matrix H, H + =(H T H) -1 H T , the training is now completed.

2. The method for constructing a sound classification model according to claim 1, characterized in that: The preprocessing includes converting the original sound data into a wav format, using a filter to eliminate environmental noise in each sound sample, cutting out the noise and silence parts in the sound waveform, and retaining the characteristic waveform.

3. The method for constructing a sound classification model according to claim 1, characterized in that: The method for pre-emphasis of the sound signal is to use a high-pass filter to enhance the high-frequency part of the sound signal, wherein the high-pass filter is: y(n)=x(n)-α*x(n-1) (1) Wherein, 0.9≤α≤1.0, n is the number of sampling points of the signal, x(n) is the sound signal, and y(n) is the sound signal after pre-emphasis.

4. The method for constructing a sound classification model according to claim 3, characterized in that: The frame division method is to divide the pre-emphasized sound signal into a plurality of frames with a duration of t.

5. The method for constructing a sound classification model according to claim 4, characterized in that: The windowing method is: The signal after framing is expressed as S(n), n=0,1,…N-1, where n is the number of frames obtained after framing, and N is the size of the frame. After adding the Hamming window, it is expressed as S′(n)=S(n)*W(n), then the form of the Hamming window W(n) is, Among them, a=0.

46.

6. The method for constructing a sound classification model according to claim 5, characterized in that: The method for performing fast Fourier transform on the sound signal is to create a Hamming window matrix C, which has the same size as S(n), multiply the corresponding positions of the two matrices to obtain a windowed sound information matrix S′(n), perform FFT fast Fourier transform on the windowed sound information matrix to obtain a matrix D, take the modulus and then square the data points of each frame in the matrix D, calculate the energy spectral density E, add the energy obtained for each frame, and obtain a new matrix F, in which each element represents the sum of the energy of each frame.

7. The method for constructing a sound classification model according to claim 6, characterized in that: The method of performing Mel filter bank and discrete cosine transform to extract spectrum features is: Define a filter bank with M filters. The filter used is a triangular filter, and its frequency response is defined as: in, Spectral line index k = (1 + L).f(m) / f s , L is the number of FFT points, f(m) is the value after the calculated Mel scale is converted to frequency, f s is the sampling frequency; The logarithmic energy of each filter bank output is then calculated as, Among them, s(m) is the logarithmic energy, X(k) is the energy spectrum of the sound signal, and H m (k) is a filter bank; After the energy of different frequency bands is logarithmically transformed, it is transformed back to the time domain through discrete cosine transform, which is called Mel frequency inverse spectrum parameter. The calculation formula is as follows:

8. A sound classification method based on an extreme learning machine, characterized in that: include, An extreme learning machine classification model is obtained by training based on the method described in any one of claims 1 to 7; The sound signal to be classified is input into the extreme learning machine classification model to obtain the classification result.

9. A sound classification system based on the method for constructing a sound classification model according to any one of claims 1 to 7, characterized in that: include, A sound pickup, including a microphone and an audio amplifier circuit, for collecting sound signals; A main processing module, including a control chip embedded with a preprocessing algorithm, wherein the preprocessing algorithm is used to receive and preprocess the sound signal obtained by the sound pickup; A DSP algorithm processing module is communicatively connected with the main processing module, and includes a chip implanted with a recognition algorithm, wherein the recognition algorithm is the classification model obtained through training, and realizes recognition and classification of natural environment sounds; The display module includes a display and a drive controller, which is used to display the execution status of the process of collecting and processing sound signals, including system startup, signal processing, and recognition results; The keyboard module starts and shuts down the system through the keyboard, and controls the execution of the system through different keys on the keyboard, including displaying the steps in progress and pausing the current system operation; A power supply and crystal oscillator module, wherein the power supply voltage is 2V to 3.6V; and the crystal oscillator module is used to provide a clock signal.

Images